Publications
111 publications, 2002–2026. List generated from HAL on 11 November 2026.
2026
-
Preprint
Micro-scale air quality models require hourly traffic data on every road segment, whereas permanent counters cover only a small part of the network. A graph neural network (GNN) was developed to reconstruct hourly traffic flows on the 32,163 road segments of Paris intra-muros from the 1,573 segments equipped with a counter (4.9%). The model adapts the Graph Neural Network for Urban Interpolation (GNNUI) to a real, sparse counting network. Flows are decoded on the edges of the junction graph (one edge per segment, both directions summed), a traffic climatology serves as an offset in a zero-inflated negative-binomial likelihood, and training proceeds by masking counters. Eight-fold spatial cross-validation, with counters sharing a published series merged before splitting, distinguishes interpolation at a withheld counter in an instrumented district (R² = 0.89, error of a quarter of the mean flow) from extrapolation to a district without counters (R² = 0.74, a third of the mean flow). These scores hold on main axes where counters are located, while relative errors are about 50% on the few counted local streets. Ablations show that the climatology offset tends to help extrapolation. Road attributes, time cycles and the live state of the counting network matter most, while weather inputs have no significant effect. The masking rate is the only inherited hyperparameter with a clear effect. When evaluated on January–August 2026 (outside the training period), the frozen model stayed around the error band of previous years.
@unpublished{jurado:hal-05856603, TITLE = {{Reconstructing hourly traffic on every street of a capital from a sparse counting network: a graph neural network applied to Paris}}, AUTHOR = {Jurado, Xavier and Wemmert, C{\'e}dric and Reiminger, Nicolas}, URL = {https://hal.science/hal-05856603}, NOTE = {working paper or preprint}, YEAR = {2026}, MONTH = Nov, KEYWORDS = {traffic reconstruction ; graph neural network ; spatial interpolation ; counting network ; cross-validation ; traffic reconstruction graph neural network spatial interpolation counting network cross-validation}, PDF = {https://hal.science/hal-05856603v1/file/manuscript.pdf}, HAL_ID = {hal-05856603}, HAL_VERSION = {v1}, } -
A taxonomy of interpretable neurosymbolic models for time series classification
Journal Knowledge-Based Systems, 2026, 352, pp.116906
A C T Time series classification (TSC) with deep learning has advanced markedly over the last decade, yet the mechanisms by which models identify and exploit discriminative patterns remain only partially understood, limiting interpretability. This paper surveys neurosymbolic approaches that aim either (i) to derive knowledge, by extracting human-readable rules and explanations from data, or (ii) to use knowledge, by injecting logical, physical, or structural constraints into learning. We review methods spanning Signal Temporal Logic (STL) formula extraction and temporal decision trees, neurosymbolic constraint synthesis and circuits, loss-based regularization including physics-informed objectives, probabilistic logic for event reasoning (e.g., DeepProbLog), and concept-level continual learning and distillation. We position these methods along common axes, explicit vs. implicit knowledge, applicability to TSC, capacity for knowledge discovery, prior knowledge requirements, and reproducibility, and discuss typical failure modes, including data bias and reasoning shortcuts. We also relate recent frugality-oriented designs that leverage non-trainable filters without sacrificing accuracy. Our analysis indicates that neurosymbolic techniques can provide actionable explanations and stronger inductive bias, with especially strong results on binary tasks, but face challenges in scaling to multi-class settings, handling shape-dependent phenomena, and ensuring transparent selection of learned rules. The survey offers a taxonomy and practical guidance for integrating symbolic structure with deep learning in TSC, highlighting open problems and promising directions.
@article{porcher:hal-05740108, TITLE = {{A taxonomy of interpretable neurosymbolic models for time series classification}}, AUTHOR = {Porcher, Damien and Ayadi, Ali and Devanne, Maxime and Weber, Jonathan and Wemmert, C{\'e}dric}, URL = {https://univoak.hal.science/hal-05740108}, JOURNAL = {{Knowledge-Based Systems}}, PUBLISHER = {{Elsevier}}, VOLUME = {352}, PAGES = {116906}, YEAR = {2026}, MONTH = Oct, DOI = {10.1016/j.knosys.2026.116906}, KEYWORDS = {Neurosymbolic learning ; Time series ; Symbolic reasoning ; Knowledge distillation ; Temporal logic ; Interpretability ; Continual learning}, HAL_ID = {hal-05740108}, HAL_VERSION = {v1}, } -
Mapping Trail Networks From Crowdsourced Hiking GPS Traces
Journal Transactions in GIS, 2026, 30 (7)
Reconstructing trail networks from GPS data is a challenging task due to the heterogeneity, sparsity, and noise inherent in crowdsourced traces. To address the dichotomy between maintaining topological connectivity and handling severe geometric noise, we present two complementary approaches for automatic trail map generation from large‐scale GPS datasets. The first is an unsupervised multigrid method that aggregates traces using a grid‐based representation, specifically addressing the preservation of structural connectivity without requiring annotated data. The second is a supervised deep learning approach based on the HRNet architecture, trained with automatically generated annotations derived from OpenStreetMap (OSM), designed to filter erratic GPS behaviors and handle complex spatial variations. Both methods are evaluated on real‐world hiking data provided by Visorando, alongside multiple benchmark datasets. We also propose dedicated evaluation metrics, including a novel intersection F ‐score (). Quantitatively, the multigrid method demonstrates high accuracy and connectivity preservation on structured or moderately noisy datasets (achieving an of 0.82 on Blaesheim and up to 0.89 on baseline pedestrian data). However, in highly complex, unconstrained mountain environments, the supervised HRNet model demonstrates superior robustness, reaching an of 0.520 and a topological intersection score () of 0.805 when trained with OSM‐augmented data. Overall, our results highlight the specific trade‐offs of each approach, providing scalable solutions for automatic trail mapping in natural environments.
@article{bletonpascal:hal-05773532, TITLE = {{Mapping Trail Networks From Crowdsourced Hiking GPS Traces}}, AUTHOR = {Bleton-Pascal, Artus and Schirm, Olivier and Weber, Jonathan and Devanne, Maxime and Lecus, Arnaud and Forestier, Germain and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-05773532}, JOURNAL = {{Transactions in GIS}}, PUBLISHER = {{Wiley}}, VOLUME = {30}, NUMBER = {7}, YEAR = {2026}, MONTH = Sep, DOI = {10.1111/tgis.70418}, KEYWORDS = {hiking map generation ; GPS data ; deep learning}, PDF = {https://hal.science/hal-05773532v1/file/Transactions%20in%20GIS%20-%202026%20-%20Bleton%E2%80%90Pascal%20-%20Mapping%20Trail%20Networks%20From%20Crowdsourced%20Hiking%20GPS%20Traces.pdf}, HAL_ID = {hal-05773532}, HAL_VERSION = {v1}, } -
RODO: A modular ontology for integrative knowledge representation in rare orodental diseases
Conference 30th International Conference on Knowledge-Based and Intelligent Information & Engineering Systems, KES International, Sep 2026, Dublin (Ireland), Ireland
Rare orodental diseases affect millions globally, yet diagnosis remains a significant challenge, with patients often waiting four to five years for a confirmed result. This diagnostic odyssey is further complicated by the fragmentation of clinical information across heterogeneous data modalities, including dental imaging, genetic profiles, and phenotypic descriptions. While general biomedical ontologies exist, they lack the domain-specific granularity and multi-axis integration required for precise orodental differential diagnosis. To address this gap, we present RODO, a modular ontology for rare orodental diseases. RODO integrates fragmented clinical phenotypes, molecular etiology, and multimodal imaging data within a single, cohesive inferential environment. We demonstrate RODO's inferential power through property chains and SWRL rules that automate severity stratification and flag potential diagnostic anomalies, among other capabilities. Logical evaluation using the Pellet reasoner confirms the ontology's consistency, while clinical use cases validate its utility for computer-aided diagnosis.
@inproceedings{wakaonyando:hal-05642091, TITLE = {{RODO: A modular ontology for integrative knowledge representation in rare orodental diseases}}, AUTHOR = {Waka Onyando, Socrates and Ayadi, Ali and Kawczynski, Marzena and Wemmert, C{\'e}dric and Bloch-Zupan, Agn{\`e}s}, URL = {https://hal.science/hal-05642091}, BOOKTITLE = {{Procedia Computer Science}}, ADDRESS = {Dublin (Ireland), Ireland}, ORGANIZATION = {{KES International}}, YEAR = {2026}, MONTH = Sep, KEYWORDS = {semantic inference ; Rare orodental diseases ; biomedical ontologies ; knowledge integration ; reification ; clinical decision support}, PDF = {https://hal.science/hal-05642091v1/file/KES_RODO__A_modular_ontology_for_integrative_knowledge_representation_in_rare_orodental_diseases.pdf}, HAL_ID = {hal-05642091}, HAL_VERSION = {v1}, } -
Transformer-Based Prediction of Conversion Yield in Catalytic Methane Decomposition
Conference 22nd IFIP WG 12.5 International Conference AIAI 2026, Jul 2026, Chania (Crète), Greece. pp.450-463
Catalytic methane decomposition (CMD) is a promising process for the simultaneous production of high-purity turquoise hydrogen and valuable solid carbon without direct carbon dioxide emissions. In particular, the CMD process utilizing carbon-based catalysts under contactless induction heating has been extensively investigated due to its operational advantages, especially the strong reaction-catalyst synergy. Despite its significant potential, optimizing this reaction remains challenging because methane conversion is highly dependent on several nonlinear and interdependent factors, including catalyst characteristics, temperature, and gas flow rate. Consequently, this study proposes a timeseries forecasting model to predict methane conversion dynamically during the reaction. Given the complex interdependencies governing the CMD process, multidimensional inputs are required, comprising endogenous, exogenous, and static variables. Therefore, a Transformer-based architecture was employed and specifically tailored for conversion forecasting. The proposed model was trained and evaluated using empirical data obtained from laboratory-scale experiments. The results demonstrate the robust potential of machine learning to predict methane conversion accurately, which is essential for real-time reaction monitoring and performance optimization. Furthermore, the model exhibits a highly promising capability to generalize and predict conversion rates when a previously unseen catalyst is introduced to the system.
@inproceedings{nguyen:hal-05697895, TITLE = {{Transformer-Based Prediction of~Conversion Yield in~Catalytic Methane Decomposition}}, AUTHOR = {Nguyen, Thi Huong Ly and Reiminger, Nicolas and Pham-Huu, Cuong and Essyed, Ahmed and Wemmert, C{\'e}dric and Jurado, Xavier}, URL = {https://hal.science/hal-05697895}, BOOKTITLE = {{IFIP Advances in Information and Communication Technology}}, ADDRESS = {Chania (Cr{\`e}te), Greece}, PUBLISHER = {{Springer Nature Switzerland}}, SERIES = {IFIP Advances in Information and Communication Technology}, VOLUME = {794}, PAGES = {450-463}, YEAR = {2026}, MONTH = Jul, DOI = {10.1007/978-3-032-30805-4\_32}, KEYWORDS = {Catalytic methane decomposition ; Time series forecasting ; Transformer models}, HAL_ID = {hal-05697895}, HAL_VERSION = {v1}, } -
Conference Advances in Intelligent Data Analysis XXIV, Apr 2026, Leiden (NL), France. pp.213-224
@inproceedings{ouaaziz:hal-05603814, TITLE = {{Guided Feature Distillation for~YOLO11: Efficient Detection of~Pseudomonas Aeruginosa in~High-Resolution Microscopy}}, AUTHOR = {Ouaaziz, Mouhcine and Klockenbring, Dimitri and Lam-Weil, Joseph and Wemmert, C{\'e}dric and Naegel, Beno{\^i}t and Chouchene, Sarah and Madec, Morgan and R{\"o}mer, Winfried}, URL = {https://hal.science/hal-05603814}, BOOKTITLE = {{Advances in Intelligent Data Analysis XXIV}}, ADDRESS = {Leiden (NL), France}, PUBLISHER = {{Springer Nature Switzerland}}, SERIES = {Lecture Notes in Computer Science}, VOLUME = {16513}, PAGES = {213-224}, YEAR = {2026}, MONTH = Apr, DOI = {10.1007/978-3-032-23833-7\_16}, HAL_ID = {hal-05603814}, HAL_VERSION = {v1}, } -
Journal Sustainable Cities and Society, 2026, 137, pp.107135
This study examines how far solar irradiations modify the wind velocity-concentration relationship commonly used in isothermal computational fluid dynamics (CFD) modeling of urban air quality. The main aim is to evaluate the validity under non-isothermal conditions of this widely used relationship and to provide new insights into the influence of solar-induced thermal effects on urban pollutant dispersion. While this relationship enables long-term concentration estimates through extrapolation from a limited set of simulations-thus offering strong operational advantages-its validity under non-isothermal conditions remains untestedYet, recent regulatory changes and empirical evidence increasingly highlight the limitations of the isothermal assumption, especially in capturing short-term pollutant dynamics influenced by solar-driven thermal effects. To address this gap, a systematic CFD analysis of pollutant dispersion within an idealized 5×5 urban building array was conducted. This array was exposed to varying inlet wind velocities and solar irradiance levels, under fixed solar position and thermal boundary conditions. Results reveal that thermally induced flow
@article{reiminger:hal-05455423, TITLE = {{Breaking the isothermal assumption in CFD air quality modeling: Solar irradiance effects on the wind velocity-concentration relationship}}, AUTHOR = {Reiminger, Nicolas and Wemmert, C{\'e}dric and Maurer, Lo{\"i}c and Vazquez, Jos{\'e} and Jurado, Xavier}, URL = {https://hal.science/hal-05455423}, JOURNAL = {{Sustainable Cities and Society}}, PUBLISHER = {{Elsevier}}, VOLUME = {137}, PAGES = {107135}, YEAR = {2026}, MONTH = Feb, DOI = {10.1016/j.scs.2026.107135}, KEYWORDS = {CFD ; air pollution ; micro-scale modeling ; wind-velocity relationship ; solar radiation ; thermal effects}, PDF = {https://hal.science/hal-05455423v1/file/Breaking%20the%20Isothermal%20Assumption%20in%20CFD%20Air%20Quality%20Modeling%20-%20accepted.pdf}, HAL_ID = {hal-05455423}, HAL_VERSION = {v1}, }
2025
-
Ontology-Guided Prompting for Reasoning in Multimodal Vision-Language Models
Conference Workshop on Multimodal Knowledge and Language Modeling at IJCAI, Aug 2025, Montréal (Québec), Canada
@inproceedings{elgohary:hal-05117925, TITLE = {{Ontology-Guided Prompting for Reasoning in Multimodal Vision-Language Models}}, AUTHOR = {Elgohary, Kareem and Ayadi, Ali and Kawczynski, Marzena and Bloch-Zupan, Agn{\`e}s and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-05117925}, BOOKTITLE = {{Workshop on Multimodal Knowledge and Language Modeling at IJCAI}}, ADDRESS = {Montr{\'e}al (Qu{\'e}bec), Canada}, YEAR = {2025}, MONTH = Aug, HAL_ID = {hal-05117925}, HAL_VERSION = {v1}, } -
Conference ORASIS 2025, ISEN Yncréa Ouest, Jun 2025, Le Croisic, France in French
Transfert de coloration assisté par MALDI-MSI : proposition d'une approche multimodale pour l'analyse histopathologique
@inproceedings{rmouque:hal-05131289, TITLE = {{Transfert de coloration assist{\'e} par MALDI-MSI : proposition d'une approche multimodale pour l'analyse histopathologique}}, AUTHOR = {Rmouque, Ilias and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-05131289}, BOOKTITLE = {{ORASIS 2025}}, ADDRESS = {Le Croisic, France}, ORGANIZATION = {{ISEN Yncr{\'e}a Ouest}}, HAL_LOCAL_REFERENCE = {105090}, YEAR = {2025}, MONTH = Jun, KEYWORDS = {histologie ; CycleGAN ; transfer ; stain ; MSI ; MALDI}, PDF = {https://hal.science/hal-05131289v1/file/Orasis_Rmouque_Ilias.pdf}, HAL_ID = {hal-05131289}, HAL_VERSION = {v1}, } -
Automated Coding of French Emergency Calls Using BERT
Conference IEEE International Symposium on Computer-Based Medical Systems (CBMS), Jun 2025, Madrid, Spain. pp.220-225
Emergency Medical Dispatch Centres (EMDC) in France are facing a growing number of calls, each requiring a rapid and efficient response, leaving no time for emergency medical dispatchers to code case files. However, this coding is essential for improving patient care through epidemiological analysis and, ultimately, for developing a decision support system. This study explores the use and deployment of natural language processing (NLP) models to automatically code records in French, based on the notes taken by call takers during emergency calls. Our work investigates fine-tuned biomedical pre-trained encoder models models for classifying highly noisy medical records, demonstrating that these models can successfully accomplish this task with a high F1 score (94%) without requiring more computationally expensive language models. To ensure the longterm reliability of the models, validation methods specifically addressing the challenges posed by EMDC data were developed. NLP can automate the coding of EMDC call records with a high degree of accuracy. However, the models currently rely solely on the written notes in the files. Future work incorporating call transcripts could further enhance their accuracy.
@inproceedings{lebot:hal-05054610, TITLE = {{Automated Coding of French Emergency Calls Using BERT}}, AUTHOR = {Lebot, Samuel and Muller, Joris and Harscoat, S{\'e}bastien and Bernhard, Delphine and Forestier, Germain and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-05054610}, BOOKTITLE = {{IEEE International Symposium on Computer-Based Medical Systems (CBMS)}}, ADDRESS = {Madrid, Spain}, PUBLISHER = {{IEEE}}, PAGES = {220-225}, YEAR = {2025}, MONTH = Jun, DOI = {10.1109/CBMS65348.2025.00053}, PDF = {https://hal.science/hal-05054610v1/file/cbms2025.pdf}, HAL_ID = {hal-05054610}, HAL_VERSION = {v1}, } -
Poster European Congress on Digital Pathology, Jun 2025, Barcelone, Spain. 2025
@misc{schaadt:hal-05080705, TITLE = {{Software and seminar format to jointly teach computational pathology in biomedical and computational academic curricula}}, AUTHOR = {Schaadt, Nadine and Neeman, Jannes and Forestier, Germain and Allerkamp, Dennis and Sprengel, Frauke and Wemmert, C{\'e}dric and Deserno, Thomas and Feuerhake, Friedrich}, URL = {https://hal.science/hal-05080705}, NOTE = {Poster}, HOWPUBLISHED = {{European Congress on Digital Pathology}}, YEAR = {2025}, MONTH = Jun, HAL_ID = {hal-05080705}, HAL_VERSION = {v1}, } -
Journal Flow Measurement and Instrumentation, 2025, 105, pp.102946
@article{zidaoui:hal-05074232, TITLE = {{Application of ResNet and Autoencoder models for anomaly detection in wastewater networks data: A Comparative study of supervised and unsupervised approaches}}, AUTHOR = {Zidaoui, Imane and Wemmert, C{\'e}dric and Joannis, Claude and Isel, Sandra and Wertel, Jonathan and Vazquez, Jos{\'e}}, URL = {https://hal.science/hal-05074232}, JOURNAL = {{Flow Measurement and Instrumentation}}, PUBLISHER = {{Elsevier}}, VOLUME = {105}, PAGES = {102946}, YEAR = {2025}, MONTH = May, DOI = {10.1016/j.flowmeasinst.2025.102946}, PDF = {https://hal.science/hal-05074232v1/file/1-s2.0-S0955598625001384-main.pdf}, HAL_ID = {hal-05074232}, HAL_VERSION = {v1}, } -
Concentration Averaging and Long-Term Exposure Assessment methodologies in Urban Areas
Journal Science of the Total Environment, 2025, 973, pp.179099
@article{jurado:hal-04987193, TITLE = {{Concentration Averaging and Long-Term Exposure Assessment methodologies in Urban Areas}}, AUTHOR = {Jurado, Xavier and Wemmert, C{\'e}dric and Maurer, Lo{\"i}c and Vazquez, Jos{\'e} and Reiminger, Nicolas}, URL = {https://hal.science/hal-04987193}, JOURNAL = {{Science of the Total Environment}}, PUBLISHER = {{Elsevier}}, VOLUME = {973}, PAGES = {179099}, YEAR = {2025}, MONTH = Mar, DOI = {10.1016/j.scitotenv.2025.179099}, HAL_ID = {hal-04987193}, HAL_VERSION = {v1}, } -
Deep Neural Network Architectures for Advanced Hiking Map Generation
Conference International Conference on Pattern Recognition Applications and Methods (ICPRAM), Feb 2025, Porto, Portugal. pp.531-541
The automation of hiking map generation using deep learning represents a pivotal advancement in geospatial analysis. This study investigates the application of neural network architectures to derive accurate hiking maps from GPS trajectory data, exclusively collected via the Visorando mobile application. By exploring the utility of 17 distinct raster features derived from geospatial data, we identify the heatmap as the most effective input for mapping intricate trail networks, achieving superior performance across accuracy, segmentation, and connectivity metrics. Among various architectures evaluated, HRNet emerged as the most efficient model, demonstrating exceptional results when combined with optimal input features, significantly outperforming state-of-the-art approaches in intersection detection and trail segmentation. This research introduces a novel framework for converting vector-based GPS traces into rasterized data suitable for convolutional neural networks, overcoming chall enges like noisy inputs and terrain variability. The findings establish a new benchmark for efficiency and accuracy in hiking map generation, offering valuable insights for the broader field of automated map inference reliant on GPS data. Future work will explore direct spatio-temporal processing of vector data, eliminating raster conversion for enhanced scalability and precision.
@inproceedings{schirm:hal-04909516, TITLE = {{Deep Neural Network Architectures for Advanced Hiking Map Generation}}, AUTHOR = {Schirm, Olivier and Devanne, Maxime and Weber, Jonathan and Lecus, Arnaud and Forestier, Germain and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04909516}, BOOKTITLE = {{International Conference on Pattern Recognition Applications and Methods (ICPRAM)}}, ADDRESS = {Porto, Portugal}, PUBLISHER = {{SCITEPRESS - Science and Technology Publications}}, PAGES = {531-541}, YEAR = {2025}, MONTH = Feb, DOI = {10.5220/0013351700003905}, PDF = {https://hal.science/hal-04909516v1/file/icpram2025.pdf}, HAL_ID = {hal-04909516}, HAL_VERSION = {v1}, } -
Conference 25ème conférence Extraction et Gestion des Connaissances (EGC), Jan 2025, Strasbourg, France
@inproceedings{ayadi:hal-04799463, TITLE = {{Hybrid ontology-based Constraint Satisfaction for object recognition in high-resolution satellite image}}, AUTHOR = {Ayadi, Zouhayra and Ayadi, Ali and Wemmert, C{\'e}dric and Teisseire, Maguelonne}, URL = {https://hal.science/hal-04799463}, BOOKTITLE = {{25{\`e}me conf{\'e}rence Extraction et Gestion des Connaissances (EGC)}}, ADDRESS = {Strasbourg, France}, YEAR = {2025}, MONTH = Jan, HAL_ID = {hal-04799463}, HAL_VERSION = {v1}, } -
Journal TSM. Techniques Sciences Méthodes, 2025 in French
@article{weber:hal-05306807, TITLE = {{Am{\'e}lioration de la pr{\'e}vision des crues par l'utilisation de r{\'e}seaux de neurones r{\'e}currents de type LSTM : {\'e}tude de cas du bassin versant de l'Ill et de la Sarre}}, AUTHOR = {Weber, Lucie and Jurado, Xavier and Nguyen, Ly and Maurer, Lo{\"i}c and Reiminger, Eva and Wemmert, C{\'e}dric and Reiminger, Nicolas}, URL = {https://univoak.hal.science/hal-05306807}, JOURNAL = {{TSM. Techniques Sciences M{\'e}thodes}}, PUBLISHER = {{Association g{\'e}n{\'e}rale des hygi{\'e}nistes et techniciens municipaux [1986-2003] - Association scientifique et technique pour l'eau et l'environnement (ASTEE) [2004-....]}}, YEAR = {2025}, HAL_ID = {hal-05306807}, HAL_VERSION = {v1}, }
2024
-
Journal Atmospheric Pollution Research, 2024, 16 (2), pp.102383
@article{reiminger:hal-04839443, TITLE = {{Advancing urban air quality modeling with solar radiation-included computational fluid dynamics simulations}}, AUTHOR = {Reiminger, Nicolas and Jurado, Xavier and Maurer, Lo{\"i}c and Vazquez, Jos{\'e} and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04839443}, JOURNAL = {{Atmospheric Pollution Research}}, PUBLISHER = {{Elsevier}}, VOLUME = {16}, NUMBER = {2}, PAGES = {102383}, YEAR = {2024}, MONTH = Dec, DOI = {10.1016/j.apr.2024.102383}, HAL_ID = {hal-04839443}, HAL_VERSION = {v1}, } -
Explaining LSTM neural networks: A DEVS-based approach
Conference International Multidisciplinary Modeling and Simulation Multiconference, Tenerife, Spain, septembre 2024, Sep 2024, Tenerife, Spain
@inproceedings{ayadi:hal-04704503, TITLE = {{Explaining LSTM neural networks: A DEVS-based approach}}, AUTHOR = {Ayadi, Ali and Truong, Thi-My-Duyen and Frydman, Claudia and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04704503}, BOOKTITLE = {{International Multidisciplinary Modeling and Simulation Multiconference, Tenerife, Spain, septembre 2024}}, ADDRESS = {Tenerife, Spain}, YEAR = {2024}, MONTH = Sep, HAL_ID = {hal-04704503}, HAL_VERSION = {v1}, } -
Journal Hand Surgery and Rehabilitation, 2024, 43 (4), pp.101742
Perilunate dislocations are a rare but serious pathology, often undetected in the emergency setting. In this study, a Deep Learning algorithm is proposed to automatically detect perilunate dislocations in frontal radiographs. A total of 374 annotated frontal wrist radiographs, comprising 345 normal and 29 pathological ones from adolescents and adults aged 16 and above with skeletal maturity, were utilized to train, validate, and test two YOLOv8 deep neural models. The first model is responsible for detecting the carpal region, and the second for segmenting a region between Gilula’s 2nd and 3rd arcs. The output of the segmentation model, trained multiple times with varying random augmentations, is then given a probability to be normal or pathological through ensemble averaging. On the considered dataset, the proposed algorithm achieves an overall F1-score of 0.880. The F1-score reaches 0.928 on the normal subgroup with a precision of 1.0, and 0.833 on the pathological subgroup with a recall (or sensitivity) of 1.0, demonstrating that the diagnosis of perilunate dislocations can be improved through automatic analysis of frontal radiographs.
@article{majzoubi:hal-04602786, TITLE = {{A Deep Learning-based Algorithm for Automatic Detection of Perilunate Dislocations in Frontal Wrist Radiographs}}, AUTHOR = {Majzoubi, Negin and All{\`e}gre, R{\'e}mi and Wemmert, C{\'e}dric and Liverneaux, Philippe}, URL = {https://hal.science/hal-04602786}, JOURNAL = {{Hand Surgery and Rehabilitation}}, PUBLISHER = {{Elsevier}}, VOLUME = {43}, NUMBER = {4}, PAGES = {101742}, YEAR = {2024}, MONTH = Jun, DOI = {10.1016/j.hansur.2024.101742}, KEYWORDS = {Medical imaging ; Computer-aided diagnosis ; Deep Learning ; Perilunate dislocations}, PDF = {https://hal.science/hal-04602786v1/file/islandora_171133.pdf}, HAL_ID = {hal-04602786}, HAL_VERSION = {v1}, } -
Multi-scale component trees for enhanced representation in multiplex immunohistochemistry imaging
Conference International Symposium on Biomedical Imaging (ISBI), May 2024, Athènes, Greece
Multiplex immunohistochemistry (mIHC) is an innovative method that simultaneously labels multiple biomarkers in the same tissue section with different colored stains. However, analyzing these complex images is a challenging task for current image processing methods. In order to efficiently process these images, it may be beneficial to employ an effective representation of the data. The component-tree is known to facilitate the representation of images containing sparse objects, which necessitate representation at a higher scale, as opposed to large background regions that can be represented at alower scale without loss of fidelity. In this paper, we present how a multi-scale version of the component-tree can be effectively utilized to represent this type of data, drawing inspiration from the approach a pathologist might take when analyzing these images (initially identifying objects of interest via a specific nuclear channel, such as DAPI, and then associating each object with a spectral signature based on the intensities of other channels). We demonstrate this method on an unsupervised analysis of 9-channel multiplexed glioblastoma data.
@inproceedings{perrin:hal-04450142, TITLE = {{Multi-scale component trees for enhanced representation in multiplex immunohistochemistry imaging}}, AUTHOR = {Perrin, Romain and Leborgne, Aur{\'e}lie and Passat, Nicolas and Naegel, Beno{\^i}t and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04450142}, BOOKTITLE = {{International Symposium on Biomedical Imaging (ISBI)}}, ADDRESS = {Ath{\`e}nes, Greece}, YEAR = {2024}, MONTH = May, DOI = {10.1109/ISBI56570.2024.10635272}, KEYWORDS = {Multi-scale ; Component-tree ; Multiplex immunohistochemistry images}, PDF = {https://hal.science/hal-04450142v1/file/Perrin_ISBI_2024.pdf}, HAL_ID = {hal-04450142}, HAL_VERSION = {v1}, } -
Multi-scale component-tree: A hierarchical representation of sparse objects
Conference International Conference on Discrete Geometry and Mathematical Morphology (DGMM), Apr 2024, Florence, Italy. pp.312-324
Component-trees are hierarchical structures developed in the framework of mathematical morphology. They model images via the inclusion relationships between the connected components of their successive threshold sets. There exist many variants of component-trees, but to the best of our knowledge, none of them deals with the representation of the image at different scales. In this article, we propose such a variant of component-tree that tackles this issue, namely the Multi-Scale Component-Tree (MSCT). We describe an algorithmic scheme for building the MSCT from the standard computation of component-trees of the image, seen from its lowest to its highest scale. At each step, a local upscaling is performed on relevant parts of the image, corresponding to nodes of the MSCT which are selected according to a stability analysis. The last step builds elements which are part of the standard component-tree (at the highest, native scale of the image). The MSCT provides a compact, efficient representation of images compared to the standard (single-scale) component-tree. In particular, the MSCT is especially suited to analyse images containing sparse objects, which require to be represented at a high scale, vs. large background regions that can be losslessly represented at a lower scale. We illustrate the relevance of the MSCT in the context of cellular image segmentation.
@inproceedings{perrin:hal-04363690, TITLE = {{Multi-scale component-tree: A hierarchical representation of sparse objects}}, AUTHOR = {Perrin, Romain and Leborgne, Aur{\'e}lie and Passat, Nicolas and Naegel, Beno{\^i}t and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04363690}, BOOKTITLE = {{Lecture Notes in Computer Science}}, ADDRESS = {Florence, Italy}, VOLUME = {14605}, PAGES = {312--324}, YEAR = {2024}, MONTH = Apr, DOI = {10.1007/978-3-031-57793-2\_24}, KEYWORDS = {mathematical morphology ; segmentation ; hierarchical representation ; component-tree ; multi-scale ; mathematical morphology}, PDF = {https://hal.science/hal-04363690v1/file/Perrin_DGMM_2024.pdf}, HAL_ID = {hal-04363690}, HAL_VERSION = {v1}, } -
Poster Colloque Français d'Intelligence Artificielle en Imagerie Biomédicale (IABM), Mar 2024, Grenoble, France in French
@misc{perrin:hal-04440939, TITLE = {{L'arbre des coupes multi-{\'e}chelles pour une meilleure repr{\'e}sentation dans l'imagerie immunohistochimique multiplex}}, AUTHOR = {Perrin, Romain and Leborgne, Aur{\'e}lie and Passat, Nicolas and Naegel, Beno{\^i}t and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04440939}, NOTE = {Poster}, HOWPUBLISHED = {{Colloque Fran{\c c}ais d'Intelligence Artificielle en Imagerie Biom{\'e}dicale (IABM)}}, YEAR = {2024}, MONTH = Mar, PDF = {https://hal.science/hal-04440939v1/file/Romain_PERRIN_IABM_2024.pdf}, HAL_ID = {hal-04440939}, HAL_VERSION = {v1}, } -
Journal Sustainable Cities and Society, 2024, in press
@article{reiminger:hal-04471929, TITLE = {{Modeling NO2 concentrations in real urban areas using computational fluid dynamics: A comparative analysis of methods to assess NO2 concentrations from NOx dispersion results}}, AUTHOR = {Reiminger, Nicolas and Jurado, Xavier and Maurer, Lo{\"i}c and Vazquez, Jos{\'e} and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04471929}, JOURNAL = {{Sustainable Cities and Society}}, PUBLISHER = {{Elsevier}}, VOLUME = {in press}, YEAR = {2024}, MONTH = Feb, DOI = {10.1016/j.scs.2024.105286}, HAL_ID = {hal-04471929}, HAL_VERSION = {v1}, } -
Journal TSM. Techniques Sciences Méthodes, 2024, 11, pp.53-71 in French
Les inondations, une aléa récurrente pouvant se montrer dévastatrice, peuvent causer de lourds dommages humains et économiques, et sont susceptibles de s'intensifier avec le changement climatique. Leurs impacts soulignent la nécessité d'anticiper les crues en développant constamment des outils efficaces et à la pointe pour la gestion du risque d'inondation. En France, la prévision des crues repose sur des outils de modélisation spécifiques basés sur la physique, la donnée ou une combinaison des deux. L'intelligence artificielle (IA), en plein essor, offre de nouvelles possibilités et perspectives pour les ingénieurs et chercheurs du domaine, mais n'est cependant pas encore utilisée à l'échelle nationale pour la prévision des crues. Bien que les algorithmes complexes d'apprentissage automatique (ML, Machine Learning) soient prometteurs pour la prévision des crues, leur mise en œuvre et leur utilisation courante restent difficiles pour les non-spécialistes. Cet article explore l'utilisation de six algorithmes d'apprentissage automatique, incluant des modèles simples d'ensemble, pour prévoir les hauteurs d'eau à 24 heures au niveau de deux stations de prévision des crues en France situées dans le Grand Est. Les résultats montrent que les niveaux d'eau, et donc les crues, peuvent être anticipés 24 heures à l'avance en utilisant uniquement des données de hauteur d'eau et par l'intermédiaire de modèles d'apprentissage automatique simples appartenant aux modèles ensemblistes. (Random Forest et Gradient Boosting). Ces derniers se révèlent performants, tout en restant accessible aux non-spécialistes, tandis que les modèles plus complexes ne permettent pas d'obtenir des résultats supérieurs en l'absence de données météorologiques supplémentaires telles que les pluies. Ce travail ouvre la voie vers l'utilisation de l'apprentissage automatique et de l'intelligence artificielle pour la prévision des crues en France. Abstract Les inondations, un risque récurrent au potentiel dévastateur, peuvent causer des dommages humains et économiques importants et sont susceptibles de s'intensifier avec le changement climatique. Leurs impacts soulignent la nécessité d'anticiper les inondations en développant continuellement des outils efficaces et de pointe pour la gestion du risque d'inondation. En France, la prévision des inondations s'appuie sur des outils de modélisation spécifiques basés sur la physique, les données ou une combinaison des deux. L'intelligence artificielle (IA), en pleine expansion, offre de nouvelles possibilités et perspectives aux ingénieurs et chercheurs du domaine, mais n'est pas encore utilisée à l'échelle nationale pour la prévision des inondations. Si les algorithmes complexes d'apprentissage automatique sont prometteurs pour la prévision des inondations, leur mise en œuvre et leur utilisation généralisée restent difficiles pour les non-spécialistes. Cet article explore l'utilisation de six algorithmes d'apprentissage automatique, dont des modèles d'ensemble simples, pour prévoir les niveaux d'eau 24 heures à l'avance dans deux stations de prévision des inondations dans la région du Grand Est en France. Les résultats démontrent que les niveaux d'eau, et par conséquent les inondations, peuvent être anticipés en utilisant uniquement des données de niveau d'eau grâce à des modèles d'apprentissage automatique simples appartenant à des modèles d'ensemble (Random Forest et Gradient Boosting). Ces modèles sont performants tout en restant accessibles aux non-spécialistes, alors que les modèles plus complexes ne donnent pas de résultats supérieurs en l’absence de données météorologiques complémentaires telles que les précipitations. Ces travaux ouvrent la voie à l’utilisation de l’apprentissage automatique et de l’intelligence artificielle dans la prévision des inondations en France.
@article{reiminger:hal-04799457, TITLE = {{{\'E}valuation des performances de l'intelligence artificielle et de l'apprentissage automatique pour la pr{\'e}vision des crues : {\'e}tude de cas du bassin versant de l'Ill}}, AUTHOR = {Reiminger, Nicolas and Jurado, Xavier and Saunier, Lo{\"i}c and Maurer, Lo{\"i}c and Reiminger, Eva and Weber, Lucie and Nguyen, Ly and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04799457}, JOURNAL = {{TSM. Techniques Sciences M{\'e}thodes}}, PUBLISHER = {{Association g{\'e}n{\'e}rale des hygi{\'e}nistes et techniciens municipaux [1986-2003] - Association scientifique et technique pour l'eau et l'environnement (ASTEE) [2004-....]}}, VOLUME = {11}, PAGES = {53-71}, YEAR = {2024}, DOI = {10.36904/tsm/20241153}, KEYWORDS = {Pr{\'e}vision, Crue, Inondation, hauteur d'eau, Apprentissage automatique, intelligence artificielle}, HAL_ID = {hal-04799457}, HAL_VERSION = {v1}, }
2023
-
Comparison of deep learning architectures for colon cancer mutation detection
Conference IEEE International Symposium on Computer Based Medical Systems (CBMS), Jun 2023, L'Aquila, Italy
Colorectal cancer is responsible of the death of hundred of thousands of people worldwide each year. The histopathological features of the tumor are generally identified from the analysis of tissue taken from a biopsy providing information for selecting the adequate treatment. With the advent of digital pathology, slides of tissue are increasingly available as Whole Slide Images (WSI) allowing their analysis using artificial intelligence algorithms. Additionally, genetic sequencing of the tumor can also be performed to identify specific mutations (e.g. KRAS) and their subtypes (e.g. G12C). While sequencing can take time and is costly, it provides key information on the specific type of cancer. In this paper, we target the identification of key genes and variants using only stained histopathology images using deep neural networks. Predicting gene information only from histopathology images could be easily used routinely and would allow saving the time and cost of tumor sequencing. Experiments performed on 45 colorectal cancer WSI from 45 patients revealed that deep models can correctly classify 90% of patients with the KRAS G12C mutation. Furthermore, we also used CAM and LIME methods to explore the interpretability of the results highlighting which parts of the images were used by the models to predict the presence of specific mutations.
@inproceedings{heckenauer:hal-04168891, TITLE = {{Comparison of deep learning architectures for colon cancer mutation detection}}, AUTHOR = {Heckenauer, Robin and Weber, Jonathan and Wemmert, C{\'e}dric and Truntzer, Caroline and Derang{\`e}re, Valentin and Ghiringhelli, Fran{\c c}ois and Hassenforder, Michel and Muller, Pierre-Alain and Forestier, Germain}, URL = {https://hal.science/hal-04168891}, BOOKTITLE = {{2023 IEEE 36th International Symposium on Computer-Based Medical Systems (CBMS)}}, ADDRESS = {L'Aquila, Italy}, PUBLISHER = {{IEEE}}, SERIES = {2023 IEEE 36th International Symposium on Computer-Based Medical Systems (CBMS)}, YEAR = {2023}, MONTH = Jun, DOI = {10.1109/CBMS58004.2023.00244}, KEYWORDS = {Digital pathology ; Histopathological images ; Colorectal cancer ; Mutation classification ; Deep learning ; Explainability}, PDF = {https://hal.science/hal-04168891v1/file/CBMS_2023_Heckenauer.pdf}, HAL_ID = {hal-04168891}, HAL_VERSION = {v1}, } -
Journal Knowledge-Based Systems, 2023, 277
Virtual stain transfer is a promising area of research in Computational Pathology, which has a great potential to alleviate important limitations when applying deep-learning-based solutions such as lack of annotations and sensitivity to a domain shift. However, in the literature, the majority of virtual staining approaches are trained for a specific staining or stain combination, and their extension to unseen stainings requires the acquisition of additional data and training. In this paper, we propose HistoStarGAN, a unified framework that performs stain transfer between multiple stainings, stain normalisation and stain invariant segmentation, all in one inference of the model. We demonstrate the generalisation abilities of the proposed solution to perform diverse stain transfer and accurate stain invariant segmentation over numerous unseen stainings, which is the first such demonstration in the field. Moreover, the pre-trained HistoStarGAN model can serve as a synthetic data generator, which paves the way for the use of fully annotated synthetic image data to improve the training of deep learning-based algorithms. To illustrate the capabilities of our approach, as well as the potential risks in the microscopy domain, inspired by applications in natural images, we generated KidneyArtPathology, a fully annotated artificial image dataset for renal pathology.
@article{vasiljevic:hal-04169986, TITLE = {{HistoStarGAN: A Unified Approach to Stain Normalisation, Stain Transfer and Stain Invariant Segmentation in Renal Histopathology}}, AUTHOR = {Vasiljevic, Jelica and Feuerhake, Friedrich and Wemmert, C{\'e}dric and Lampert, Thomas}, URL = {https://hal.science/hal-04169986}, JOURNAL = {{Knowledge-Based Systems}}, PUBLISHER = {{Elsevier}}, VOLUME = {277}, YEAR = {2023}, MONTH = Jun, DOI = {10.1016/j.knosys.2023.110780}, PDF = {https://hal.science/hal-04169986v1/file/S0950705123005300.pdf}, HAL_ID = {hal-04169986}, HAL_VERSION = {v1}, } -
Conference ORASIS 2023 : journées francophones des jeunes chercheurs en vision par ordinateur, May 2023, Carqueiranne, France in French
Le cancer colorectal est responsable de la mort de centaines de milliers de personnes dans le monde chaque année. Les caractéristiques histopathologiques de la tumeur sont généralement identifiées à partir de l'analyse des tissus prélevés lors d'une biopsie, ce qui permet de choisir le traitement adéquat. Avec l'avènement de la pathologie numérique, les lames de tissu sont de plus en plus disponibles sous forme d'images de lames histopathologiques (WSI), ce qui permet leur analyse par des algorithmes d'intelligence artificielle. En outre, le séquençage génétique de la tumeur peut également être effectué pour identifier des mutations spécifiques (e.g. KRAS) et leurs sous-types (e.g. G12C). Bien que le séquençage prenne du temps et soit coûteux, il fournit des informations essentielles sur le type spécifique de cancer. Dans cet article, nous ciblons l'identification de gènes et de variants clés en utilisant uniquement des images histopathologiques à l'aide de réseaux de neurones profonds. La prédiction d'informations génétiques uniquement à partir d'images histopathologiques pourrait être facilement utilisée en routine et permettrait d'économiser le temps et le coût du séquençage des tumeurs. Les expériences réalisées sur 45 WSI de cancer colorectal provenant de 45 patients ont révélé que les modèles profonds peuvent classer correctement 90% des patients présentant la mutation G12C du gène KRAS.En outre, nous avons également utilisé les méthodes CAM et LIME pour explorer l'interprétabilité des résultats en mettant en évidence les parties des images utilisées par les modèles pour prédire la présence de mutations spécifiques.
@inproceedings{heckenauer:hal-04185535, TITLE = {{Comparaison des architectures d'apprentissage profond pour la d{\'e}tection des mutations du cancer du c{\^o}lon}}, AUTHOR = {Heckenauer, R and Weber, J and Wemmert, C{\'e}dric and Truntzer, C and Derang{\`e}re, V and Ghiringhelli, F and Hassenforder, Michel and Muller, P-A and Forestier, Germain}, URL = {https://hal.science/hal-04185535}, BOOKTITLE = {{ORASIS 2023 : journ{\'e}es francophones des jeunes chercheurs en vision par ordinateur,}}, ADDRESS = {Carqueiranne, France}, VOLUME = {2023}, YEAR = {2023}, MONTH = May, KEYWORDS = {explicabilit{\'e} ; apprentissage profond ; classification des mutations ; cancer colorectal ; images histopathologiques ; Pathologie num{\'e}rique}, PDF = {https://hal.science/hal-04185535v1/file/ORASIS_2023_Heckenauer.pdf}, HAL_ID = {hal-04185535}, HAL_VERSION = {v1}, } -
Conference ORASIS 2023, Laboratoire LIS, UMR 7020, May 2023, Carqueiranne, France in French
Comparaison des architectures d'apprentissage profond pour la détection des mutations du cancer du côlon
@inproceedings{heckenauer:hal-04219453, TITLE = {{Comparaison des architectures d'apprentissage profond pour la d{\'e}tection des mutations du cancer du c{\^o}lon}}, AUTHOR = {Heckenauer, Robin and Weber, Jonathan and Wemmert, C{\'e}dric and Truntzer, Caroline and Derang{\`e}re, Valentin and Ghiringhelli, Fran{\c c}ois and Hassenforder, Michel and Muller, Pierre-Alain and Forestier, Germain}, URL = {https://hal.science/hal-04219453}, BOOKTITLE = {{ORASIS 2023}}, ADDRESS = {Carqueiranne, France}, ORGANIZATION = {{Laboratoire LIS, UMR 7020}}, HAL_LOCAL_REFERENCE = {83935}, YEAR = {2023}, MONTH = May, PDF = {https://hal.science/hal-04219453v1/file/ORASIS_2023_2.pdf}, HAL_ID = {hal-04219453}, HAL_VERSION = {v1}, } -
Conference ORASIS 2023, Laboratoire LIS, UMR 7020, May 2023, Carqueiranne, France in French
Une approche multi-grilles pour la génération automatique d'une carte de sentiers de randonnée à partir de données GPS
@inproceedings{schirm:hal-04219570, TITLE = {{Une approche multi-grilles pour la g{\'e}n{\'e}ration automatique d'une carte de sentiers de randonn{\'e}e {\`a} partir de donn{\'e}es GPS}}, AUTHOR = {Schirm, Olivier and Devanne, Maxime and Weber, Jonathan and Lecus, Arnaud and Forestier, Germain and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04219570}, BOOKTITLE = {{ORASIS 2023}}, ADDRESS = {Carqueiranne, France}, ORGANIZATION = {{Laboratoire LIS, UMR 7020}}, HAL_LOCAL_REFERENCE = {75793}, YEAR = {2023}, MONTH = May, PDF = {https://hal.science/hal-04219570v1/file/496934.pdf}, HAL_ID = {hal-04219570}, HAL_VERSION = {v1}, } -
Journal Sustainable Cities and Society, 2023, 99
Urban areas face a significant health risk due to atmospheric pollution necessitating continuous monitoring to assess and mitigate its impact and achieve a sustainable city. In this study, a deep learning model was developed and trained using Computational Fluid Dynamics (CFD) simulations to predict the dispersion of nitrogen dioxide () emissions from traffic. The model’s performance was evaluated by comparing it to real-world field measurements conducted in Antwerp, Belgium, throughout 2016. Temporal comparisons with a traffic-influenced air quality station over the entire year yielded an average correlation coefficient (R) of 0.83 and a mean relative error (MRE) of 0.21. Additionally, a spatial evaluation was conducted by comparing the model’s predictions to a measurement campaign involving 73 samplers. The spatial evaluation resulted in an R of 0.72 and a MRE of 0.18 for samplers located near known emission sources. Notably, the deep learning model demonstrated computational efficiency, outperforming CFD simulations by an order of magnitude of 100 to 1000, enabling real-time pollution monitoring and large-scale scenario studies without compromising the consideration of micro-scale effects that are typically overlooked in larger-scale models.
@article{jurado:hal-04217618, TITLE = {{Assessment of a Deep Learning Model for Monitoring Atmospheric Pollution: Case Study in Antwerp, Belgium}}, AUTHOR = {Jurado, Xavier and Reiminger, Nicolas and Maurer, Lo{\"i}c and Vazquez, Jos{\'e} and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04217618}, JOURNAL = {{Sustainable Cities and Society}}, PUBLISHER = {{Elsevier}}, VOLUME = {99}, YEAR = {2023}, DOI = {10.1016/j.scs.2023.104951}, KEYWORDS = {Deep learning ; Convolutional neural network ; Computational fluid dynamics ; Air quality ; Microscale modelling}, HAL_ID = {hal-04217618}, HAL_VERSION = {v1}, } -
Journal Journal of Wind Engineering and Industrial Aerodynamics, 2023, 235, pp.105361
@article{reiminger:hal-04012575, TITLE = {{Influence of depressed road configuration on downwind pollutant concentrations: A CFD study under various thermal stability conditions}}, AUTHOR = {Reiminger, Nicolas and Jurado, Xavier and Maurer, Lo{\"i}c and Vazquez, Jos{\'e} and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04012575}, JOURNAL = {{Journal of Wind Engineering and Industrial Aerodynamics}}, PUBLISHER = {{Elsevier}}, VOLUME = {235}, PAGES = {105361}, YEAR = {2023}, DOI = {10.1016/j.jweia.2023.105361}, PDF = {https://hal.science/hal-04012575v1/file/S0167610523000648.pdf}, HAL_ID = {hal-04012575}, HAL_VERSION = {v1}, } -
Journal Atmosphere, 2023, 14 (2)
@article{jurado:hal-03998101, TITLE = {{On the Correlations between Particulate Matter: Comparison between Annual/Monthly Concentrations and PM10/PM2.5}}, AUTHOR = {Jurado, Xavier and Reiminger, Nicolas and Maurer, Lo{\"i}c and Vazquez, Jos{\'e} and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-03998101}, JOURNAL = {{Atmosphere}}, PUBLISHER = {{MDPI }}, VOLUME = {14}, NUMBER = {2}, YEAR = {2023}, DOI = {10.3390/atmos14020385}, PDF = {https://hal.science/hal-03998101v1/file/islandora_159683.pdf}, HAL_ID = {hal-03998101}, HAL_VERSION = {v1}, } -
Journal Water Science and Technology, In press, 87 (12)
To prevent the pollution of water resources, the measurement and the limitation of wastewater discharges are required. Despite the progress in the field of data acquisition systems, sensors are subject to malfunctions that can bias the evaluation of the pollution flow. It is therefore essential to identify potential anomalies in the data before any use. The objective of this work is to deploy artificial intelligence tools to automate the data validation and to assess the added value of this approach in assisting the validation performed by an operator. To do so, we compare two state-of-the-art anomaly detection algorithms on turbidity data in a sewer network. On the one hand, we conclude that the One-class SVM model is not adapted to the nature of the studied data which is heterogeneous and noisy. The Matrix Profile model, on the other hand, provides promising results with a majority of anomalies detected and a relatively limited number of false positives. By comparing these results to the expert validation, it turns out that the use of the Matrix Profile model objectifies and accelerates the validation task while maintaining the same level of performance compared to the annotator agreement rate between two experts.
@article{zidaoui:hal-04116074, TITLE = {{Validation of wastewater data using artificial intelligence tools and the evaluation of their performance regarding annotator agreement}}, AUTHOR = {Zidaoui, Imane and Wemmert, C{\'e}dric and Dufresne, Matthieu and Joannis, Claude and Isel, Sandra and Wertel, Jonathan and Vazquez, Jos{\'e}}, URL = {https://hal.science/hal-04116074}, JOURNAL = {{Water Science and Technology}}, PUBLISHER = {{IWA Publishing}}, VOLUME = {87}, NUMBER = {12}, YEAR = {2023}, DOI = {10.2166/wst.2023.174}, KEYWORDS = {annotator agreement ; artificial intelligence ; data validation ; matrix profile ; one-class SVM ; wastewater}, PDF = {https://hal.science/hal-04116074v1/file/wst2023174.pdf}, HAL_ID = {hal-04116074}, HAL_VERSION = {v1}, }
2022
-
Conditional image synthesis for improved segmentation of glomeruli in renal histopathological images
Conference IEEE-EMBS International Conference on Biomedical and Health Informatics, Sept. 27-30 2022, Ioannina, Greece, Sep 2022, Ioannina, Greece
@inproceedings{allender:hal-03797648, TITLE = {{Conditional image synthesis for improved segmentation of glomeruli in renal histopathological images}}, AUTHOR = {Allender, Florian and All{\`e}gre, R{\'e}mi and Wemmert, C{\'e}dric and Dischler, Jean-Michel}, URL = {https://hal.science/hal-03797648}, BOOKTITLE = {{IEEE-EMBS International Conference on Biomedical and Health Informatics, Sept. 27-30 2022, Ioannina, Greece}}, ADDRESS = {Ioannina, Greece}, YEAR = {2022}, MONTH = Sep, DOI = {10.1109/BHI56158.2022.9926880}, PDF = {https://hal.science/hal-03797648v1/file/allender%20et%20al.pdf}, HAL_ID = {hal-03797648}, HAL_VERSION = {v1}, } -
Méthodes de classification du cancer du sein pour les microscopes à réalité augmentée
Conference RFIAP 2022 (Congrès Reconnaissance des Formes, Image, Apprentissage et Perception), Jul 2022, Vannes, France in French
Un microscope à réalité augmentée (ARM) permet d'afficher des informations supplémentaires sur les tissus en cours d'analyse par un pathologiste. Les méthodes d'analyse d'image utilisées par ces microscopes doivent être robustes aux changements de niveau de grossissement pour suivre les déplacements dans les lames de verres. Nous résolvons ce problème en tirant parti des caractéristiques présentes dans certains niveaux de grossissement clefs qui permettent de mieux comprendre d'autres niveaux de grossissement. Nous proposons une méthode temps réel robuste aux changements de niveaux de grossissement en entraînant des réseaux de neurones profonds sur certains niveaux clefs d'images de lames (WSI) et en les testant sur les niveaux d'un microscope. Nous montrons que notre approche surpasse ou équivaut les méthodes naïves sur un jeu de données de cancer du sein pour l'architecture d'apprentissage profond Inception-ResNet-v2.
@inproceedings{heckenauer:hal-03687423, TITLE = {{M{\'e}thodes de classification du cancer du sein pour les microscopes {\`a} r{\'e}alit{\'e} augment{\'e}e}}, AUTHOR = {Heckenauer, Robin and Weber, Jonathan and Wemmert, C{\'e}dric and Hassenforder, Michel and Muller, Pierre-Alain and Forestier, Germain}, URL = {https://hal.science/hal-03687423}, BOOKTITLE = {{RFIAP 2022 (Congr{\`e}s Reconnaissance des Formes, Image, Apprentissage et Perception)}}, ADDRESS = {Vannes, France}, YEAR = {2022}, MONTH = Jul, KEYWORDS = {microscope {\`a} r{\'e}alit{\'e} augment{\'e}e ; r{\'e}seaux convolutifs profonds ; pathologie num{\'e}rique ; Classification en temps r{\'e}el du cancer du sein}, PDF = {https://hal.science/hal-03687423v1/file/Robin_Heckenauer_RFIAP_2022_Camera_Ready.pdf}, HAL_ID = {hal-03687423}, HAL_VERSION = {v1}, } -
Generation of synthetic training data for SEEG electrodes segmentation
Journal International Journal of Computer Assisted Radiology and Surgery, 2022, 17 (5), pp.937-943
@article{pantovic:hal-03752554, TITLE = {{Generation of synthetic training data for SEEG electrodes segmentation}}, AUTHOR = {Pantovic, Anja and Ren, Xiaoxi and Wemmert, C{\'e}dric and Ollivier, Ir{\`e}ne and Essert, Caroline}, URL = {https://hal.science/hal-03752554}, JOURNAL = {{International Journal of Computer Assisted Radiology and Surgery}}, PUBLISHER = {{Springer Verlag}}, VOLUME = {17}, NUMBER = {5}, PAGES = {937--943}, YEAR = {2022}, MONTH = Mar, DOI = {10.1007/s11548-022-02585-4}, PDF = {https://hal.science/hal-03752554v1/file/Generation%20of%20synthetic%20training%20data%20for%20SEEG%20electrodes%20segmentation.pdf}, HAL_ID = {hal-03752554}, HAL_VERSION = {v1}, } -
Ensemble clustering for histopathological images segmentation using convolutional autoencoders
Conference International Conference on Computer Vision Theory and Applications (VISAPP), Feb 2022, En ligne, France. pp.933-940
Unsupervised deep learning using autoencoders has shown excellent results in image analysis and computer vision. However, only few studies have been presented in the field of digital pathology, where proper labelling of the objects of interest is a particularly costly and difficult task. Thus, having a first fully unsupervised segmentation could greatly help in the analysis process of such images. In this paper, many architectures of convolutional autoencoders have been compared to study the influence of three main hyperparameters: (1) number of convolutional layers, (2) number of convolutions in each layer and (3) size of the latent space. Different clustering algorithms are also compared and we propose a new way to obtain more precise results by applying ensemble clustering techniques which consists in combining multiple clustering results.
@inproceedings{rmouque:hal-03469780, TITLE = {{Ensemble clustering for histopathological images segmentation using convolutional autoencoders}}, AUTHOR = {Rmouque, Ilias and Weber, Jonathan and Devanne, Maxime and Forestier, Germain and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-03469780}, BOOKTITLE = {{International Conference on Computer Vision Theory and Applications (VISAPP)}}, ADDRESS = {En ligne, France}, PAGES = {933-940}, YEAR = {2022}, MONTH = Feb, DOI = {10.5220/0010835300003124}, PDF = {https://hal.science/hal-03469780v1/file/visapp2022.pdf}, HAL_ID = {hal-03469780}, HAL_VERSION = {v1}, } -
CycleGAN for virtual stain transfer: is seeing really believing?
Journal Artificial Intelligence in Medicine, 2022, 133, pp.102420
@article{vasiljevic:hal-03799506, TITLE = {{CycleGAN for virtual stain transfer: is seeing really believing?}}, AUTHOR = {Vasiljevic, Jelica and Nisar, Zeeshan and Feuerhake, Friedrich and Wemmert, C{\'e}dric and Lampert, Thomas}, URL = {https://hal.science/hal-03799506}, JOURNAL = {{Artificial Intelligence in Medicine}}, PUBLISHER = {{Elsevier}}, VOLUME = {133}, PAGES = {102420}, YEAR = {2022}, DOI = {10.1016/j.artmed.2022.102420}, PDF = {https://hal.science/hal-03799506v1/file/Vasiljevic%20et%20al.pdf}, HAL_ID = {hal-03799506}, HAL_VERSION = {v1}, } -
Journal Computer Methods and Programs in Biomedicine, 2022, 221
@article{allender:hal-03736936, TITLE = {{Data augmentation based on spatial deformations for histopathology: An evaluation in the context of glomeruli segmentation}}, AUTHOR = {Allender, Florian and All{\`e}gre, R{\'e}mi and Wemmert, C{\'e}dric and Dischler, Jean-Michel}, URL = {https://hal.science/hal-03736936}, JOURNAL = {{Computer Methods and Programs in Biomedicine}}, PUBLISHER = {{Elsevier}}, VOLUME = {221}, YEAR = {2022}, DOI = {10.1016/j.cmpb.2022.106919}, PDF = {https://hal.science/hal-03736936v1/file/allender%20et%20al.pdf}, HAL_ID = {hal-03736936}, HAL_VERSION = {v1}, } -
Deep Learning methods evaluation to predict air quality based on computational fluid dynamics
Journal Expert Systems with Applications, 2022, 203
@article{jurado:hal-03736929, TITLE = {{Deep Learning methods evaluation to predict air quality based on computational fluid dynamics}}, AUTHOR = {Jurado, Xavier and Reiminger, Nicolas and Benmoussa, Marouane and Vazquez, Jos{\'e} and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-03736929}, JOURNAL = {{Expert Systems with Applications}}, PUBLISHER = {{Elsevier}}, VOLUME = {203}, YEAR = {2022}, DOI = {10.1016/j.eswa.2022.117294}, PDF = {https://hal.science/hal-03736929v1/file/S0957417422006595.pdf}, HAL_ID = {hal-03736929}, HAL_VERSION = {v1}, } -
Journal TSM. Techniques Sciences Méthodes, 2022, 11, pp.39-51 in French
Pour lutter contre la pollution des masses d’eaux, la réglementation française impose la mesure et la régulation des rejets d’eaux usées dans l’environnement. Cependant, malgré les progrès dans le domaine des systèmes d’acquisition de données, les capteurs, tout particulièrement les sondes de turbidité, installés dans des milieux agressifs tels que les réseaux d’assainissement sont sujets à des dysfonctionnements fréquents (dérive, saturation, données manquantes…), qui peuvent fausser l’évaluation du flux de pollution. Il est donc essentiel d’identifier les potentielles anomalies avant toute utilisation des données. Aujourd’hui, cette validation se fait au niveau de la supervision et/ou via des opérateurs. L’objectif de ce travail est d’évaluer le potentiel des outils d’intelligence artificielle à automatiser la validation et d’estimer la plus-value de cette approche par rapport à une validation « métier » effectuée par un expert. Pour cela, quatre algorithmes de détection d’anomalies de l’état de l’art sont comparés en utilisant des données de turbidité issues du réseau de collecte de Saint-Malo Agglomération. La plupart de ces algorithmes ne sont pas adaptés à la nature des données étudiées qui sont hétérogènes et bruitées. Seul l’algorithme Matrix Profile permet d’obtenir des résultats prometteurs avec une majorité d’anomalies détectées et un nombre de faux positifs relativement limités.
@article{zidaoui:hal-04463350, TITLE = {{Utilisation de l'intelligence artificielle pour la validation des mesures en continu de la pollution des eaux us{\'e}es}}, AUTHOR = {Zidaoui, Imane and Joannis, Claude and Wertel, Jonathan and Isel, Sandra and Wemmert, C{\'e}dric and Vazquez, Jos{\'e} and Dufresne, Matthieu}, URL = {https://hal.science/hal-04463350}, JOURNAL = {{TSM. Techniques Sciences M{\'e}thodes}}, PUBLISHER = {{Association g{\'e}n{\'e}rale des hygi{\'e}nistes et techniciens municipaux [1986-2003] - Association scientifique et technique pour l'eau et l'environnement (ASTEE) [2004-....]}}, NUMBER = {11}, PAGES = {39-51}, YEAR = {2022}, DOI = {10.36904/tsm/202211039}, KEYWORDS = {anomaly ; artificial intelligence ; data analysis ; pollution ; flow rate monitoring ; overflow monitoring ; wastewater networks ; Anomalie ; Intelligence artificielle ; Analyse de donn{\'e}es ; Pollution ; Diagnostic permanent ; Autosurveillance ; Assainissement}, HAL_ID = {hal-04463350}, HAL_VERSION = {v1}, } -
Journal Artificial Intelligence in Medicine, 2022
Recently, Artificial Intelligence namely Deep Learning methods have revolutionized a wide range of domains and applications. Besides, Digital Pathology has so far played a major role in the diagnosis and the prognosis of tumors. However, the characteristics of the Whole Slide Images namely the gigapixel size, high resolution and the shortage of richly labeled samples have hindered the efficiency of classical Machine Learning methods. That goes without saying that traditional methods are poor in generalization to different tasks and data contents. Regarding the success of Deep learning when dealing with Large Scale applications, we have resorted to the use of such models for histopathological image segmentation tasks. First, we review and compare the classical UNet and Att-UNet models for colon cancer WSI segmentation in a sparsely annotated data scenario. Then, we introduce novel enhanced models of the Att-UNet where different schemes are proposed for the skip connections and spatial attention gates positions in the network. In fact, spatial attention gates assist the training process and enable the model to avoid irrelevant feature learning. Alternating the presence of such modules namely in our Alter-AttUNet model adds robustness and ensures better image segmentation results. In order to cope with the lack of richly annotated data in our AiCOLO colon cancer dataset, we suggest the use of a multi-step training strategy that also deals with the WSI sparse annotations and unbalanced class issues. All proposed methods outperform state-of-the-art approaches but Alter-AttUNet generates the best compromise between accurate results and light network. The model achieves 95.88% accuracy with our sparse AiCOLO colon cancer datasets. Finally, to evaluate and validate our proposed architectures we resort to publicly available WSI data: the NCT-CRC-HE-100K, the CRC-5000 and the Warwick colon cancer histopathological dataset. Respective accuracies of 99.65%, 99.73% and 79.03% were reached. A comparison with state-of-art approaches is established to view and compare the key solutions for histopathological image segmentation.
@article{benhamida:hal-03797652, TITLE = {{Weakly Supervised Learning using Attention gates for colon cancer histopathological image segmentation}}, AUTHOR = {Ben Hamida, Amina and Devanne, Maxime and Weber, Jonathan and Truntzer, Caroline and Derang{\`e}re, Valentin and Ghiringhelli, Fran{\c c}ois and Forestier, Germain and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-03797652}, JOURNAL = {{Artificial Intelligence in Medicine}}, PUBLISHER = {{Elsevier}}, YEAR = {2022}, DOI = {10.1016/j.artmed.2022.102407}, PDF = {https://hal.science/hal-03797652v1/file/aim2022.pdf}, HAL_ID = {hal-03797652}, HAL_VERSION = {v1}, }
2021
-
Détection en temps réel des glomérules en pathologie rénale
Conference ORASIS 2021, Centre National de la Recherche Scientifique [CNRS], Sep 2021, Saint Ferréol, France in French
Le domaine de la pathologie numérique a émergé avec l'introduction des scanners de lames et a conduit au développement de nouveaux outils d'analyse de lames histopathologiques. La disponibilité de la représentation numérique des lames a motivé le développement de méthodes issues de l'intelligence artificielle pour identifier automatiquement les structures microscopiques afin d'accompagner les pathologistes dans leur diagnostic. Contrairement à de nombreuses approches existantes visant la détection de structures microscopiques sur des images statiques à un niveau de grossissement donné et fixe, nos travaux se concentrent sur la détection en temps réel des structures à différentes échelles. En effet, la détection en temps réel à différentes échelles apporte des défis supplémentaires, mais imite également mieux la façon dont les pathologistes travaillent, car ils déplacent continuellement les lames et changent le niveau de grossissement pendant leur analyse. Dans cet article, nous nous concentrons sur la pathologie rénale et plus particulièrement sur la détection en temps réel des glomérules à différentes échelles. Notre méthode est basée sur le modèle de détection d'objets d'apprentissage profond YOLOv3 pré-entraîné sur le jeu de données COCO et affiné pour détecter les glomérules. Nous étudions les avantages de l'utilisation d'images multi-échelles pour améliorer la capacité du réseau à détecter les glomérules à des niveaux de grossissement variables en temps réel.
@inproceedings{heckenauer:hal-03339642, TITLE = {{D{\'e}tection en temps r{\'e}el des glom{\'e}rules en pathologie r{\'e}nale}}, AUTHOR = {Heckenauer, Robin and Weber, Jonathan and Wemmert, C{\'e}dric and Feuerhake, Friedrich and Hassenforder, Michel and Muller, Pierre-Alain and Forestier, Germain}, URL = {https://hal.science/hal-03339642}, BOOKTITLE = {{ORASIS 2021}}, ADDRESS = {Saint Ferr{\'e}ol, France}, ORGANIZATION = {{Centre National de la Recherche Scientifique [CNRS]}}, YEAR = {2021}, MONTH = Sep, KEYWORDS = {r{\'e}seaux convolutifs profonds ; pathologie num{\'e}rique ; d{\'e}tection de glom{\'e}rule en temps r{\'e}el}, PDF = {https://hal.science/hal-03339642v1/file/ORASIS2021.pdf}, HAL_ID = {hal-03339642}, HAL_VERSION = {v1}, } -
Journal Sustainable Cities and Society, 2021, 71
Computational fluid dynamics has shown a great interest among the scientific community to assess air pollutant concentrations in urban areas and, define strategies to limit air pollution and achieve sustainable cities of the future. Recent studies have given methodologies on how to assess mean annual concentrations based on numerical model results to compare with the annual air quality standards. Nonetheless, these methodologies need many wind directions to be modelled and, therefore, lead to high calculation costs. The purpose of this paper is to present two approaches to decrease the calculation costs when calculating annual concentration from computational fluid dynamics results by (1) ignoring uniformly spaced wind direction and (2) considering the predominant wind directions. According to the results, the first approach is on overall better than the second one for any wind rose or building layout considered. With the first approach, the calculation costs can be reduced up to 50 % without leading to more than 20 % of error, and even less error can be expected for homogeneous wind roses. Finally, a method to finely evaluate errors made when using the first approach versus using the whole wind rose, without computing it, is presented.
@article{jurado:hal-03243967, TITLE = {{On the minimal wind directions required to assess mean annual air pollution concentration based on CFD results}}, AUTHOR = {Jurado, Xavier and Reiminger, Nicolas and Vazquez, Jos{\'e} and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-03243967}, JOURNAL = {{Sustainable Cities and Society}}, PUBLISHER = {{Elsevier}}, VOLUME = {71}, YEAR = {2021}, MONTH = Aug, DOI = {10.1016/j.scs.2021.102920}, PDF = {https://hal.science/hal-03243967v1/file/S2210670721002079.pdf}, HAL_ID = {hal-03243967}, HAL_VERSION = {v1}, } -
Conference Atelier Apprentissage profond: Théorie et Applications (APTA), Journées Francophones Extraction et Gestion des Connaissances (EGC), Jan 2021, Montpellier, France in French
L'apprentissage profond non-supervisé à partir d'auto-encodeurs permet d'obtenir d'excellents résultats en matière d'analyse d'images et de vision par ordinateur. Cependant, seules quelques études ont été présentées dans le domaine de la pathologie numérique, où l'étiquetage précis des objets d'intérêt est une tâche particulièrement coûteuse et difficile. Ainsi, le fait de disposer d'une première segmentation entièrement non supervisée pourrait grandement faciliter le processus d'analyse de ces images. Dans cet article, de nombreuses architectures d'auto-encodeurs convolutifs ont été comparées afin d'étudier l'influence de trois principaux paramètres : 1) le nombre de couches convolutives, 2) le nombre de convolutions dans chacune de ces couches et 3) la taille de l'espace latent. Différents algorithmes de clustering sont également comparés et une nouvelle approche permettant d'obtenir des résultats plus précis en appliquant des techniques de clustering par ensemble est présentée.
@inproceedings{rmouque:hal-03604158, TITLE = {{Segmentation non-supervis{\'e}e d'images histopathologiques {\`a} l'aide d'auto-encodeurs et d'ensemble clustering}}, AUTHOR = {Rmouque, Ilias and Devanne, Maxime and Weber, Jonathan and Forestier, Germain and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-03604158}, BOOKTITLE = {{Atelier Apprentissage Profond : Th{\'e}orie et Applications (APTA) Atelier organis{\'e} dans le cadre de la conf{\'e}rence EGC 2021}}, ADDRESS = {Montpellier, France}, YEAR = {2021}, MONTH = Jan, PDF = {https://hal.science/hal-03604158v1/file/Rmouque_APTA2021.pdf}, HAL_ID = {hal-03604158}, HAL_VERSION = {v1}, } -
Journal Computer Methods and Programs in Biomedicine, 2021, pp.106157
Objective: This article presents an automatic image processing framework to extract quantitative high-level information describing the micro-environment of glomeruli in consecutive whole slide images (WSIs) processed with different staining modalities of patients with chronic kidney rejection after kidney transplantation.Methods: This four-step framework consists of: 1) approximate rigid registration, 2) cell and anatomical structure segmentation 3) fusion of information from different stainings using a newly developed registration algorithm 4) feature extraction.Results: Each step of the framework is validated independently both quantitatively and qualitatively by pathologists. An illustration of the different types of features that can be extracted is presented.Conclusion: The proposed generic framework allows for the analysis of the micro-environment surrounding large structures that can be segmented (either manually or automatically). It is independent of the segmentation approach and is therefore applicable to a variety of biomedical research questions.Significance: Chronic tissue remodelling processes after kidney transplantation can result in interstitial fibrosis and tubular atrophy (IFTA) and glomerulosclerosis. This pipeline provides tools to quantitatively analyse, in the same spatial context, information from different consecutive WSIs and help researchers understand the complex underlying mechanisms leading to IFTA and glomerulosclerosis
@article{merveille:hal-03224927, TITLE = {{An automatic framework for Fusing Information from differently stained consecutive digital whole slide images: A case study in renal histology}}, AUTHOR = {Merveille, Odyss{\'e}e and Lampert, Thomas and Schmitz, Jessica and Forestier, Germain and Feuerhake, Friedrich and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-03224927}, JOURNAL = {{Computer Methods and Programs in Biomedicine}}, PUBLISHER = {{Elsevier}}, PAGES = {106157}, YEAR = {2021}, DOI = {10.1016/j.cmpb.2021.106157}, KEYWORDS = {digital pathology ; chromogenic duplex immunohistochemistry ; digital whole slide image ; glomeruli segmentation ; glomeruli matching ; brightfield images}, PDF = {https://hal.science/hal-03224927v1/file/CMPB_Sysmifta.pdf}, HAL_ID = {hal-03224927}, HAL_VERSION = {v1}, } -
Proceedings Upper-Rhine Artificial Intelligence Symposium 2021, Oct 2021, Kaiserslautern, France. 2021
@proceedings{jurado:hal-03321653, TITLE = {{Deep Learning associated with Computational Fluid Dynamics to predict pollution concentration fields in urban areas}}, AUTHOR = {Jurado, Xavier and Reiminger, Nicolas and Benmoussa, Marouane and Vazquez, Jos{\'e} and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-03321653}, BOOKTITLE = {{Upper-Rhine Artificial Intelligence Symposium 2021}}, ADDRESS = {Kaiserslautern, France}, YEAR = {2021}, HAL_ID = {hal-03321653}, HAL_VERSION = {v1}, } -
Deep learning for colon cancer histopathological images analysis
Journal Computers in Biology and Medicine, 2021
Nowadays, digital pathology plays a major role in the diagnosis and prognosis of tumours. Unfortunately, existing methods remain limited when faced with the high resolution and size of Whole Slide Images (WSIs) coupled with the lack of richly annotated datasets. Regarding the ability of the Deep Learning (DL) methods to cope with the large scale applications, such models seem like an appealing solution for tissue classification and segmentation in histopathological images. This paper focuses on the use of DL architectures to classify and highlight colon cancer regions in a sparsely annotated histopathological data context. First, we review and compare state-of-the-art Convolutional Neural networks (CNN) including the AlexNet, vgg, ResNet, DenseNet and Inception models. To cope with the shortage of rich WSI datasets, we have resorted to the use of transfer learning techniques. This strategy comes with the hallmark of relying on a large size computer vision dataset (ImageNet) to train the network and generate a rich collection of learnt features. The testing and evaluation of such models on our AiCOLO colon cancer dataset ensure accurate patch-level classification results reaching up to 96.98% accuracy rate with ResNet. The CNN models have also been tested and evaluated with the CRC-5000, nct-crc-he-100k and merged datasets. ResNet respectively achieves 96.77%, 99.76% and 99.98% for the three publicly available datasets. Then, we present a pixel-wise segmentation strategy for colon cancer WSIs through the use of both UNet and SegNet models. We introduce a multi-step training strategy as a remedy for the sparse annotation of histopathological images. UNet and SegNet are used and tested in different training scenarios including data augmentation and transfer learning and ensure up to 76.18% and 81.22% accuracy rates. Besides, we test our training strategy and models on the CRC-5000, nct-crc-he-100k and Warwick datasets. Respective accuracy rates of 98.66%, 99.12% and 78.39% were achieved by SegNet. Finally, we analyze the existing models to discover the most suitable network and the most effective training strategy for our colon tumour segmentation case study.
@article{benhamida:hal-03321646, TITLE = {{Deep learning for colon cancer histopathological images analysis}}, AUTHOR = {Ben Hamida, Amina and Devanne, Maxime and Weber, Jonathan and Truntzer, Caroline and Derang{\`e}re, Valentin and Ghiringhelli, Fran{\c c}ois and Forestier, Germain and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-03321646}, JOURNAL = {{Computers in Biology and Medicine}}, PUBLISHER = {{Elsevier}}, YEAR = {2021}, DOI = {10.1016/j.compbiomed.2021.104730}, KEYWORDS = {Artificial intelligence ; Transfer learning ; Semantic segmentation ; Deep learning ; Tumour segmentation ; Colon cancer ; Whole slide images ; Digital pathology}, PDF = {https://hal.science/hal-03321646v1/file/cbm2021.pdf}, HAL_ID = {hal-03321646}, HAL_VERSION = {v1}, } -
Deep Learning for Histopathological Image Analysis
Book chapter Mourad Elloumi. Deep Learning for Biomedical Data Analysis. Techniques, Approaches, and Applications, Springer, pp.153-169, 2021, 978-3-030-71675-2
Anatomical Pathology dates back to the 19th century when Rudolf Virchow introduced his concept of cellular pathology and when the technical improvements of light microscopy enabled widespread use of structural criteria to define diseases. Since then, the quality of optical instruments has been constantly evolving. However the central element of the diagnostic process remains the knowledge and experience of pathologists visually classifying observations according to internationally agreed guidelines (e.g., World Health Organisation (WHO) classification), and much of the pre-analytical steps of specimen preparation (e.g., fixation, embedding, sectioning, staining) is only partially automated and still requires many manual steps. Thanks to the recent advent and costeffectiveness of digital scanners, tissue histopathology slides can now be fully digitized and stored as Whole Slide Images (WSI). With the availability and analysis of a much larger set of variables combined with sophisticated imaging and analytic techniques, the traditional paradigm of pathology based on visually descriptive microscopy can be complemented and substantially improved by digital pathology, utilizing screen-based visualization of digital tissue sections and novel analysis tools potentially combining the conventional evaluation by pathologists with a computerbased diagnostic aid system. A central element of such evolving medical utilities and decision support systems will be image analysis, a field in which Deep Learning (DL) has recently made immense progress, notably the work of Lecun et al. [33] on Convolutional Neural Networks (CNNs) and especially the development of very large Artificial Neural Networks (ANNs) that are revolutionizing the field. Indeed, they have surpassed all existing image processing methods in most fields (segmentation, object detection, classification, etc.). All current methods applied to histopathological image analysis will be presented as well as the future technological issues and challenges of this discipline.
@incollection{wemmert:hal-03321639, TITLE = {{Deep Learning for Histopathological Image Analysis}}, AUTHOR = {Wemmert, C{\'e}dric and Weber, Jonathan and Feuerhake, Friedrich and Forestier, Germain}, URL = {https://hal.science/hal-03321639}, BOOKTITLE = {{Deep Learning for Biomedical Data Analysis. Techniques, Approaches, and Applications}}, EDITOR = {Mourad Elloumi}, PUBLISHER = {{Springer}}, PAGES = {153-169}, YEAR = {2021}, DOI = {10.1007/978-3-030-71676-9\_7}, KEYWORDS = {deep learning ; whole slide images ; digital pathology ; machine learning}, PDF = {https://hal.science/hal-03321639v1/file/dlfbda2021.pdf}, HAL_ID = {hal-03321639}, HAL_VERSION = {v1}, } -
Self Adversarial Attack as an Augmentation Method for Immunohistochemical Stainings
Conference IEEE International Symposium on Biomedical Imaging, 2021, En ligne, France
It has been shown that unpaired image-to-image translation methods constrained by cycle-consistency hide the information necessary for accurate input reconstruction as imperceptible noise. We demonstrate that, when applied to histopathology data, this hidden noise appears to be related to stain specific features and show that this is the case with two immunohistochemical stainings during translation to Periodic acid-Schiff (PAS), a histochemical staining method commonly applied in renal pathology. Moreover, by perturbing this hidden information, the translation models produce different, plausible outputs. We demonstrate that this property can be used as an augmentation method which, in a case of supervised glomeruli segmentation, leads to improved performance.
@inproceedings{vasiljevic:hal-03321035, TITLE = {{Self Adversarial Attack as an Augmentation Method for Immunohistochemical Stainings}}, AUTHOR = {Vasiljevic, Jelica and Feuerhake, Friedrich and Wemmert, C{\'e}dric and Lampert, Thomas}, URL = {https://hal.science/hal-03321035}, BOOKTITLE = {{2021 IEEE International Symposium on Biomedical Imaging}}, ADDRESS = {En ligne, France}, YEAR = {2021}, DOI = {10.1109/ISBI48211.2021.9433838}, KEYWORDS = {digital pathology, image-to-image translation, cycle-consistency, self adversarial attack}, PDF = {https://hal.science/hal-03321035v1/file/islandora_117460.pdf}, HAL_ID = {hal-03321035}, HAL_VERSION = {v1}, } -
Journal Neurocomputing, 2021, 460, pp.277-291
@article{vasiljevic:hal-03321037, TITLE = {{Towards Histopathological Stain Invariance by Unsupervised Domain Augmentation using Generative Adversarial Networks}}, AUTHOR = {Vasiljevic, Jelica and Feuerhake, Friedrich and Wemmert, C{\'e}dric and Lampert, Thomas}, URL = {https://hal.science/hal-03321037}, JOURNAL = {{Neurocomputing}}, PUBLISHER = {{Elsevier}}, VOLUME = {460}, PAGES = {277-291}, YEAR = {2021}, DOI = {10.1016/j.neucom.2021.07.005}, PDF = {https://hal.science/hal-03321037v1/file/S0925231221010390.pdf}, HAL_ID = {hal-03321037}, HAL_VERSION = {v1}, }
2020
-
Preprint
Objective: This article presents an automatic image processing framework to extract quantitative high-level information describing the micro-environment of glomeruli in consecutive whole slide images (WSIs) processed with different staining modalities of patients with chronic kidney rejection after kidney transplantation. Methods: This three step framework consists of: 1) cell and anatomical structure segmentation based on colour decon-volution and deep learning 2) fusion of information from different stainings using a newly developed registration algorithm 3) feature extraction. Results: Each step of the framework is validated independently both quantitatively and qualitatively by pathologists. An illustration of the different types of features that can be extracted is presented. Conclusion: The proposed generic framework allows for the analysis of the micro-environment surrounding large structures that can be segmented (either manually or automatically). It is independent of the segmentation approach and is therefore applicable to a variety of biomedical research questions. Significance: Chronic tissue remodelling processes after kidney transplantation can result in in-terstitial fibrosis and tubular atrophy (IFTA) and glomeru-losclerosis. This pipeline provides tools to quantitatively analyse, in the same spatial context, information from different consecutive WSIs and help researchers understand the complex underlying mechanisms leading to IFTA and glomerulosclerosis.
@unpublished{merveille:hal-02925040, TITLE = {{An automatic framework to study the tissue micro-environment of renal glomeruli in differently stained consecutive digital whole slide images}}, AUTHOR = {Merveille, Odyss{\'e}e and Lampert, Thomas and Schmitz, Jessica and Forestier, Germain and Feuerhake, Friedrich and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-02925040}, NOTE = {working paper or preprint}, YEAR = {2020}, MONTH = Aug, KEYWORDS = {digital pathology ; brightfield images ; chromogenic duplex immunohistochemistry ; digital whole slide image ; glomeruli segmentation ; glomeruli matching}, PDF = {https://hal.science/hal-02925040v1/file/glomeruli_analysis_framework.pdf}, HAL_ID = {hal-02925040}, HAL_VERSION = {v1}, } -
Real-Time Detection of Glomeruli in Renal Pathology
Conference IEEE International Symposium on Computer-Based Medical Systems (CBMS), Jul 2020, Rochester (on line), United States. pp.350-355
The field of digital pathology emerged with the introduction of whole slide imaging scanners and lead to the development of new tools for analyzing histopathological slides. The availability of digital representation of the slides has motivated the development of artificial intelligence methods to automatically identify microscopic structures in order to support pathologists in their diagnosis. Unlike many existing approaches targeting the detection of microscopic structures on static images at a given and fixed magnification level, our work focuses on the real-time detection of the structures at different scales. Indeed, real-time detection at different scales brings additional challenges but also better mimics the way pathologists work as they continuously move the slides and change the magnification level during their analysis. In this paper, we focus on renal pathology and more specifically on the real-time detection of glomeruli at different scales. Our method is based on the deep learning object detection model YOLOv3 pre-trained on the COCO dataset and fine tuned to detect glomeruli. We investigate the benefits of using multiscale images to improve the network ability to detect glomeruli at variable magnification levels in real time. Index Terms-real-time glomerulus detection, digital pathology, deep convolutional networks
@inproceedings{heckenauer:hal-03353271, TITLE = {{Real-Time Detection of Glomeruli in Renal Pathology}}, AUTHOR = {Heckenauer, Robin and Weber, Jonathan and Wemmert, Cedric and Feuerhake, Friedrich and Hassenforder, Michel and Muller, Pierre-Alain and Forestier, Germain}, URL = {https://hal.science/hal-03353271}, BOOKTITLE = {{IEEE International Symposium on Computer-Based Medical Systems (CBMS)}}, ADDRESS = {Rochester (on line), United States}, PUBLISHER = {{IEEE}}, PAGES = {350-355}, YEAR = {2020}, MONTH = Jul, DOI = {10.1109/cbms49503.2020.00072}, PDF = {https://hal.science/hal-03353271v1/file/cbms2020.pdf}, HAL_ID = {hal-03353271}, HAL_VERSION = {v1}, } -
Journal Sustainable Cities and Society, 2020, 59, pp.102221
Numerical models are valuable tools to assess air pollutant concentrations in cities which can be used to define new strategies to achieve sustainable cities of the future in terms of air quality. Numerical results are however difficult to be directly compared to air quality standards since they are usually valid only for specific wind speed and direction while some standards are on annual values. The purpose of this paper is to present existing and new methodologies to turn numerical results into mean annual concentrations and discuss their limitations. To this end, methodologies to assess wind speed distribution based on wind rose data are presented first. Then, methodologies are compared to assess mean annual concentrations based on numerical results and on wind speed distributions. According to the results, a Weibull distribution can be used to accurately assess wind speed distribution in France, but the results can be improved using a sigmoid function presented in this paper. It is also shown that using the wind rose data directly to assess mean annual concentrations can lead to underestimations of annual concentrations. Finally, the limitations of discrete methodologies to assess mean annual concentrations are discussed and a new methodology using continuous functions is described.
@article{reiminger:hal-03051698, TITLE = {{Methodologies to assess mean annual air pollution concentration combining numerical results and wind roses}}, AUTHOR = {Reiminger, Nicolas and Jurado, Xavier and Vazquez, Jos{\'e} and Wemmert, C{\'e}dric and Blond, Nad{\`e}ge and Wertel, Jonathan and Dufresne, Matthieu}, URL = {https://hal.science/hal-03051698}, JOURNAL = {{Sustainable Cities and Society}}, PUBLISHER = {{Elsevier}}, VOLUME = {59}, PAGES = {102221}, YEAR = {2020}, MONTH = Jun, DOI = {10.1016/j.scs.2020.102221}, KEYWORDS = {59 ; Reiminger ; N. ; Jurado ; X. ; Vazquez ; 2020. Methodologies to assess mean annual air pollution concentration combining numerical results and wind roses. Sustainable Cities and Society ; Wertel ; Blond ; M. ; Dufresne ; C. ; Wemmert ; J.}, PDF = {https://hal.science/hal-03051698v1/file/PrePrint-Methodologieswindroses.pdf}, HAL_ID = {hal-03051698}, HAL_VERSION = {v1}, } -
Journal Journal of Wind Engineering and Industrial Aerodynamics, 2020, 200, pp.104160
@article{reiminger:hal-03041494, TITLE = {{Effects of wind speed and atmospheric stability on the air pollution reduction rate induced by noise barriers}}, AUTHOR = {Reiminger, Nicolas and Jurado, Xavier and Vazquez, Jos{\'e} and Wemmert, C{\'e}dric and Blond, Nad{\`e}ge and Dufresne, Matthieu and Wertel, Jonathan}, URL = {https://hal.science/hal-03041494}, JOURNAL = {{Journal of Wind Engineering and Industrial Aerodynamics}}, PUBLISHER = {{Elsevier}}, VOLUME = {200}, PAGES = {104160}, YEAR = {2020}, MONTH = May, DOI = {10.1016/j.jweia.2020.104160}, PDF = {https://hal.science/hal-03041494v2/file/Accepted%20manuscript%20-%20Effects%20of%20wind%20speed%20and%20atmospheric%20stability%20on%20the%20air%20pollution%20reduction%20rate%20induced%20by%20noise%20barriers.pdf}, HAL_ID = {hal-03041494}, HAL_VERSION = {v2}, } -
Journal La Revue de Médecine Interne, 2020, 41, pp.192 - 195
@article{pelaccia:hal-03489735, TITLE = {{Une intelligence artificielle raisonne-t-elle de la m{\^e}me fa{\c c}on que les cliniciens pour poser des diagnostics ?}}, AUTHOR = {Pelaccia, T. and Forestier, G. and Wemmert, C.}, URL = {https://hal.science/hal-03489735}, JOURNAL = {{La Revue de M{\'e}decine Interne}}, PUBLISHER = {{Elsevier}}, VOLUME = {41}, PAGES = {192 - 195}, YEAR = {2020}, MONTH = Mar, DOI = {10.1016/j.revmed.2019.12.014}, PDF = {https://hal.science/hal-03489735v1/file/S0248866320300023.pdf}, HAL_ID = {hal-03489735}, HAL_VERSION = {v1}, } -
Assessment of mean annual NO2 concentration based on a partial dataset
Journal Atmospheric Environment, 2020, 221, pp.117087
@article{jurado:hal-02398708, TITLE = {{Assessment of mean annual NO2 concentration based on a partial dataset}}, AUTHOR = {Jurado, Xavier and Reiminger, Nicolas and Vazquez, Jos{\'e} and Wemmert, C{\'e}dric and Dufresne, Matthieu and Blond, Nad{\`e}ge and Wertel, Jonathan}, URL = {https://hal.science/hal-02398708}, JOURNAL = {{Atmospheric Environment}}, PUBLISHER = {{Elsevier}}, VOLUME = {221}, PAGES = {117087}, YEAR = {2020}, MONTH = Jan, DOI = {10.1016/j.atmosenv.2019.117087}, PDF = {https://hal.science/hal-02398708v1/file/Accepted%20manuscript%20-%20Assessment%20of%20mean%20annual%20NO2%20concentration%20based%20on%20a%20partial%20dataset.pdf}, HAL_ID = {hal-02398708}, HAL_VERSION = {v1}, } -
GAN-based data augmentation for histopathological image segmentation
Conference Atelier Extraction et Gestion des Connaissances (EGC) - Apprentissage Profond: Théorie et Applications (APTA), Jan 2020, Bruxelles, Belgium
@inproceedings{allender:hal-05508204, TITLE = {{GAN-based data augmentation for histopathological image segmentation}}, AUTHOR = {Allender, Florian and All{\`e}gre, R{\'e}mi and Wemmert, C{\'e}dric and Dischler, Jean-Michel}, URL = {https://univoak.hal.science/hal-05508204}, BOOKTITLE = {{Atelier Extraction et Gestion des Connaissances (EGC) - Apprentissage Profond: Th{\'e}orie et Applications (APTA)}}, ADDRESS = {Bruxelles, Belgium}, YEAR = {2020}, MONTH = Jan, HAL_ID = {hal-05508204}, HAL_VERSION = {v1}, } -
Intelligence artificielle et analyse d’images histologiques : Opportunités et challenges
Journal Revue Française d'Histotechnologie, 2020, 32 (1), pp.213-238 in French
L’anatomopathologie remonte au 19ème siècle. Depuis, le matériel n’a cessé d’évoluer, mais l’oeil expert du pathologiste reste incontournable. De même, le processus complet (préparation du tissu et des lames, coloration, contrôle qualité) consiste encore en beaucoup d’étapes manuelles. Grâce à l’avènement récent de scanners numériques à bon rendement, les lames d’histopathologie tissulaire peuvent maintenant être numérisées et stockées sous forme d’images numériques. Avec la disponibilité et l’analyse d’un ensemble beaucoup plus vaste de variables combinées à des techniques d’imagerie et d’analyse sophistiquées, le paradigme traditionnel du pathologiste et de la microscopie pourrait rapidement être remplacé par la pathologie numérique, s’appuyant sur une visualisation sur écran des coupes de tissus numérisées et une analyse combinant un pathologiste et un système d’aide au diagnostic informatisé. En termes d’analyse d’images, d’immenses progrès ont aussi été faits récemment, notamment les travaux de Lecun et al. sur les réseaux de neurones convolutionnels et surtout le développement de réseaux de neurones de très grandes tailles (apprentissage profond) qui révolutionnent le domaine. En effet, ceux-ci ont surpassé l’ensemble des méthodes de traitement d’images existantes dans la plupart des domaines (segmentation, détection d’objets, classification). L’ensemble des méthodes actuelles appliquées à l’analyse d’images histopathologiques sera présenté ainsi que les enjeux et défis technologiques futurs de cette discipline.
@article{wemmert:hal-03979724, TITLE = {{Intelligence artificielle et analyse d'images histologiques : Opportunit{\'e}s et challenges}}, AUTHOR = {Wemmert, C{\'e}dric and Weber, Jonathan and Forestier, Germain}, URL = {https://hal.science/hal-03979724}, JOURNAL = {{Revue Fran{\c c}aise d'Histotechnologie}}, PUBLISHER = {{Association fran{\c c}aise d'histotechnologie}}, VOLUME = {32}, NUMBER = {1}, PAGES = {213-238}, YEAR = {2020}, DOI = {10.25830/afh.rfh.2020.32.213.238}, KEYWORDS = {deep learning ; histopathological image analysis ; analyse d'images histopathologiques ; apprentissage profond}, PDF = {https://hal.science/hal-03979724v1/file/Revue_afh_2020_12.pdf}, HAL_ID = {hal-03979724}, HAL_VERSION = {v1}, }
2019
-
Deconstructing the diagnostic reasoning of human versus artificial intelligence
Journal CMAJ, 2019, 191 (48), pp.E1332-E1335
@article{pelaccia:hal-02632899, TITLE = {{Deconstructing the diagnostic reasoning of human versus artificial intelligence}}, AUTHOR = {Pelaccia, Thierry and Forestier, Germain and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-02632899}, JOURNAL = {{CMAJ}}, PUBLISHER = {{Canadian Medical Association}}, VOLUME = {191}, NUMBER = {48}, PAGES = {E1332-E1335}, YEAR = {2019}, MONTH = Dec, DOI = {10.1503/cmaj.190506}, PDF = {https://hal.science/hal-02632899v1/file/cmaj2019.pdf}, HAL_ID = {hal-02632899}, HAL_VERSION = {v1}, } -
Journal Heritage Journal, 2019, pp.938-942
The presence of high endothelial venules (HEVs) in the tumor micro-environment is associated with favorable clinical out-come in breast cancer and other human solid tumors. HEVs can be highlighted in tissue samples by MECA-79 immunohistochemistry. HEVs are heterogeneously scattered in the tumor micro-environment, preferentially in areas infiltrated by lymphocytes, and their density must be assessed at high resolution on whole slide in the tumor area. We present a frame-work for HEV detection through classification of MECA-79 immunostaining in whole slide images. A classifier is provided, as well as a whole slide image metric for general performance evaluation of detectors. An original ensemble strategy for convolutional neural networks has been developed and its relevance in MECA-79 and general immunostaining classification tasks is discussed.
@article{abreu:hal-04771691, TITLE = {{Ensemble Of Neural Networks For High Endothelial Venules Detection In Meca-79 Immunohistochemistry Images}}, AUTHOR = {Abreu, A. and Franchet, C. and Frenois, F-X. and Brousset, P. and Girard, Jean-Philippe and Denefle, P. and Naegel, B. and Wemmert, C.}, URL = {https://ut-toulouseinp.hal.science/hal-04771691}, JOURNAL = {{Heritage Journal}}, PUBLISHER = {{The National Heritage Board, Singapore}}, PAGES = {938-942}, YEAR = {2019}, MONTH = Jul, DOI = {10.1109/isbi.2019.8759578}, KEYWORDS = {High endothelial venules ; whole slide images ; convolutional neural network ; ensemble learning}, HAL_ID = {hal-04771691}, HAL_VERSION = {v1}, } -
Crowdsourcing of Histological Image Labeling and Object Delineation by Medical Students
Journal IEEE Transactions on Medical Imaging, 2019, 38 (5), pp.1284-1294
Crowdsourcing in pathology has been performed on tasks that are assumed to be manageable by nonexperts. Demand remains high for annotations of more complex elements in digital microscopic images, such as anatomical structures. Therefore, this work investigates conditions to enable crowdsourced annotations of high-level image objects, a complex task considered to require expert knowledge. 76 medical students without specific domain knowledge who voluntarily participated in three experiments solved two relevant annotation tasks on histopathological images: (1) Labeling of images showing tissue regions, and (2) delineation of morphologically defined image objects. We focus on methods to ensure sufficient annotation quality including several tests on the required number of participants and on the correlation of participants' performance between tasks. In a set up simulating annotation of images with limited ground truth, we validated the feasibility of a confidence score using full ground truth. For this, we computed a majority vote using weighting factors based on individual assessment of contributors against scattered gold standard annotated by pathologists. In conclusion, we provide guidance for task design and quality control to enable a crowdsourced approach to obtain accurate annotations required in the era of digital pathology.
@article{grote:hal-02299285, TITLE = {{Crowdsourcing of Histological Image Labeling and Object Delineation by Medical Students}}, AUTHOR = {Grote, Anne and Schaadt, Nadine S and Forestier, Germain and Feuerhake, Friedrich and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-02299285}, JOURNAL = {{IEEE Transactions on Medical Imaging}}, PUBLISHER = {{Institute of Electrical and Electronics Engineers}}, VOLUME = {38}, NUMBER = {5}, PAGES = {1284-1294}, YEAR = {2019}, MONTH = May, DOI = {10.1109/TMI.2018.2883237}, KEYWORDS = {confidence score ; majority vote ; annotation ; digital pathology ; image delineation ; image classification ; human decision making ; Crowdsourcing}, PDF = {https://hal.science/hal-02299285v1/file/tmi2018.pdf}, HAL_ID = {hal-02299285}, HAL_VERSION = {v1}, } -
Strategies for Training Stain Invariant CNNS
Conference IEEE International Symposium on Biomedical Imaging (ISBI), Apr 2019, Venise, Italy. pp.905-909
An important part of Digital Pathology is the analysis of multiple digitised whole slide images from differently stained tissue sections. It is common practice to mount consecutive sections containing corresponding microscopic structures on glass slides, and to stain them differently to highlight specific tissue components. These multiple staining modalities result in very different images but include a significant amount of consistent image information. Deep learning approaches have recently been proposed to analyse these images in order to automatically identify objects of interest for pathologists. These supervised approaches require a vast amount of annotations, which are difficult and expensive to acquire-a problem that is multiplied with multiple stainings. This article presents several training strategies that make progress towards stain invariant networks. By training the network on one commonly used staining modality and applying it to images that include corresponding but differently stained tissue structures, the presented unsupervised strategies demonstrate significant improvements over standard training strategies.
@inproceedings{lampert:hal-02283639, TITLE = {{Strategies for Training Stain Invariant CNNS}}, AUTHOR = {Lampert, Thomas and Merveille, Odyss{\'e}e and Schmitz, Jessica and Forestier, Germain and Feuerhake, Friedrich and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-02283639}, NOTE = {2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)}, BOOKTITLE = {{2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)}}, ADDRESS = {Venise, Italy}, PUBLISHER = {{IEEE}}, SERIES = {2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)}, PAGES = {905-909}, YEAR = {2019}, MONTH = Apr, DOI = {10.1109/ISBI.2019.8759266}, KEYWORDS = {multi-stain analysis ; digital pathology ; whole slide images ; deep convolutional networks}, PDF = {https://hal.science/hal-02283639v1/file/main.pdf}, HAL_ID = {hal-02283639}, HAL_VERSION = {v1}, }
2018
-
Role of Task Complexity and Training in Crowdsourced Image Annotation
Conference MICCAI Workshop on Computational Pathology, Sep 2018, Grenade, Spain. pp.44-51
Accurate annotation of anatomical structures or pathological changes in microscopic images is an important task in computational pathology. Crowdsourcing holds promise to address this demand, but so far feasibility has only be shown for simple tasks and not for high-quality annotation of complex structures which is often limited by shortage of experts. Third-year medical students participated in solving two complex tasks, labeling of images and delineation of relevant image objects in breast cancer and kidney tissue. We evaluated their performance and addressed the requirements of task complexity and training phases. Our results show feasibility and a high agreement between students and experts. The training phase improved accuracy of image labeling.
@inproceedings{schaadt:hal-04558736, TITLE = {{Role of Task Complexity and Training in Crowdsourced Image Annotation}}, AUTHOR = {Schaadt, Nadine and Grote, Anne and Forestier, Germain and Wemmert, C{\'e}dric and Feuerhake, Friedrich}, URL = {https://uha.hal.science/hal-04558736}, BOOKTITLE = {{Computational Pathology and Ophthalmic Medical Image Analysis}}, ADDRESS = {Grenade, Spain}, PUBLISHER = {{Springer International Publishing}}, SERIES = {Lecture Notes in Computer Science}, VOLUME = {11039}, PAGES = {44-51}, YEAR = {2018}, MONTH = Sep, DOI = {10.1007/978-3-030-00949-6\_6}, HAL_ID = {hal-04558736}, HAL_VERSION = {v1}, } -
Model-based graph segmentation in 2-D fluorescence microsecopy images
Conference 2018 24th International Conference on Pattern Recognition (ICPR), Aug 2018, Beijing, France. pp.3844-3849
@inproceedings{abreu:hal-05490037, TITLE = {{Model-based graph segmentation in 2-D fluorescence microsecopy images}}, AUTHOR = {Abreu, A. and Frenoist, F.X. and Valitutti, S. and Brousset, P. and Denefle, P. and Naegel, B. and Wemmert, C.}, URL = {https://univoak.hal.science/hal-05490037}, BOOKTITLE = {{2018 24th International Conference on Pattern Recognition (ICPR)}}, ADDRESS = {Beijing, France}, PUBLISHER = {{IEEE}}, PAGES = {3844-3849}, YEAR = {2018}, MONTH = Aug, DOI = {10.1109/ICPR.2018.8546011}, HAL_ID = {hal-05490037}, HAL_VERSION = {v1}, } -
Collaborative Clustering: Why, When, What and How
Journal Information Fusion, 2018, 39, pp.81-95
Clustering is one type of unsupervised learning where the goal is to partition the set of objects into groups called clusters. Faced to the difficulty to design a general purpose clustering algorithm and to choose a good, let alone perfect, set of criteria for clustering a data set, one solution is to resort to a variety of clustering procedures based on different techniques, parameters and/or initial-izations, in order to construct one (or several) final clustering(s). The hope is that by combining several clustering solutions, each one with its own bias and imperfections, one will get a better overall solution. In the cooperative clustering model, as Ensemble Clustering, a set of clustering algorithms are used in parallel on a given data set: the local results are combined to get an hopefully better overall clustering. In the collaborative framework, the goal is that each local computation, quite possibly applied to distinct data sets, benefit from the work done by the other collaborators. This paper is dedicated to collaborative clustering. In particular, after a brief overview of clustering and the major issues linked to, it presents main challenges related to organize and control the collaborative process. http://dx.doi.org/10.1016/j.inffus.2017.04.008
@article{cornuejols:hal-02327993, TITLE = {{Collaborative Clustering: Why, When, What and How}}, AUTHOR = {Cornu{\'e}jols, Antoine and Wemmert, C{\'e}dric and Gan{\c c}arski, Pierre and Bennani, Youn{\`e}s}, URL = {https://hal.science/hal-02327993}, JOURNAL = {{Information Fusion}}, PUBLISHER = {{Elsevier}}, VOLUME = {39}, PAGES = {81-95}, YEAR = {2018}, MONTH = Jan, DOI = {10.1016/j.inffus.2017.04.008}, KEYWORDS = {Cooperative clustering ; Collaborative Clustering ; Clustering combining}, PDF = {https://hal.science/hal-02327993v1/file/Collaborative%20Clustering%20-%20Why%20When%20What%20et%20How.pdf}, HAL_ID = {hal-02327993}, HAL_VERSION = {v1}, } -
Conference Conference of the European Association of Systems Medicine: Big Data - Transition to Practice, Utrecht, Netherlands, janvier 2018, Jan 2018, Utrecht, Netherlands
@inproceedings{grote:hal-04549033, TITLE = {{Learn and Participate: Crowdsourcing of Image Annotation Supports Interactive Teaching in Histopathology and Involves Medical Students in Systems Medicine Research}}, AUTHOR = {Grote, Anne and Schaadt, Nadine S. and Forestier, Germain and Wemmert, C{\'e}dric and Feuerhake, Friederich}, URL = {https://hal.science/hal-04549033}, BOOKTITLE = {{IEEE Transactions on Medical Imaging}}, ADDRESS = {Utrecht, Netherlands}, VOLUME = {38}, NUMBER = {5}, YEAR = {2018}, MONTH = Jan, HAL_ID = {hal-04549033}, HAL_VERSION = {v1}, }
2017
-
Clustering collaboratif : Principes et mise en oeuvre
Conference BDA (Gestion de Données - Principes, Technologies et Applications), Nov 2017, Nancy, France in French
Pour tenter de faire sens des masses de données disponibles en quantité croissante, il est nécessaire de disposer d’outils performants limitant l’implication, souvent chronophage, de l’expert. Les méthodes non supervisées d’exploration de données telles que les méthodes de clustering sont une réponse à ce besoin. Cependant, leur mise en œuvre effective demande que l’utilisateur présuppose un certain nombre de propriétés des structures internes des données à mettre en évidence telles que, par exemple, leur type ou simplement leur nombre. Il doit aussi être capable de traduire cet a priorisous forme d’un ou plusieurs objectifs d’analyse et de choisir les méthodes adéquates (et leurs paramètres) de façon optimale. Malheureusement, ceci est une tâche ardue pour laquelle il n’existe pasactuellement de « recette miracle ». Par ailleurs, l’utilisation de bases de données distantes pour lesquelles il peut être intéressant de partager des informations nécessite souvent des analyses locales afin de préserver la confidentialité, .Dans ce contexte, les approches collaboratives, dans lesquelles différents algorithmes de clustering bien que travaillant sur des données éventuellement différentes, échangent et partagent des informations apparaissent comme une voie prometteuse. En effet, les expériences montrent que l’influence des choix initiaux des méthodes et de leurs paramètres est fortement atténué et qu’un tel schéma permet de découvrir de meilleures structures que si chacune de ces méthodestravaillait isolément.Cet article présente les questions que posent le clustering en général, et le clustering distribué en particulier ainsi que les défis à relever. Il propose ensuite un cadre dans lequel il est possible d’organiser et d’inscrire les approches de clustering collaboratif. Les différentes possibilités sont illustrées par des exemples de travaux existants.
@inproceedings{gancarski:hal-04338863, TITLE = {{Clustering collaboratif : Principes et mise en oeuvre}}, AUTHOR = {Gan{\c c}arski, Pierre and Cornu{\'e}jols, Antoine and Wemmert, C{\'e}dric and Bennani, Youn{\`e}s}, URL = {https://hal.science/hal-04338863}, BOOKTITLE = {{BDA (Gestion de Donn{\'e}es - Principes, Technologies et Applications)}}, ADDRESS = {Nancy, France}, YEAR = {2017}, MONTH = Nov, KEYWORDS = {Machine Learning}, PDF = {https://hal.science/hal-04338863v1/file/2017_BDA_conf_Clustering_Collaboratif.pdf}, HAL_ID = {hal-04338863}, HAL_VERSION = {v1}, } -
Detection of glomeruli in renal pathology by mutual comparison of multiple staining modalities
Conference International Symposium on Image and Signal Processing and Analysis (ISPA), Sep 2017, Ljubljana, Slovenia. pp.19-24
We evaluate the detection of glomerular structures in whole slide images (WSIs) of histopathological slides stained with multiple histochemical and immuno-histochemical staining using a convolutional neural network (CNN) based approach. We mutually compare the CNN performance on different stainings (Jones H&E, PAS, Sirius Red and CD10) and we present a novel approach to improve glomeruli detection on one staining by taking into account the classification results from differently stained consecutive sections of the same tissue. Using this integrative approach, the detection rate (F1-score) on a single stain can be improved by up to 30%.
@inproceedings{temerinacott:hal-03807632, TITLE = {{Detection of glomeruli in renal pathology by mutual comparison of multiple staining modalities}}, AUTHOR = {Temerinac-Ott, M. and Forestier, Germain and Schmitz, J. and Hermsen, M. and Brasen, J. H. and Feuerhake, F. and Wemmert, C.}, URL = {https://hal.science/hal-03807632}, BOOKTITLE = {{Proceedings of the 10th International Symposium on Image and Signal Processing and Analysis}}, ADDRESS = {Ljubljana, Slovenia}, PUBLISHER = {{IEEE}}, PAGES = {19-24}, YEAR = {2017}, MONTH = Sep, DOI = {10.1109/ispa.2017.8073562}, PDF = {https://hal.science/hal-03807632v1/file/ispa2017.pdf}, HAL_ID = {hal-03807632}, HAL_VERSION = {v1}, } -
Optimal cut in minimum spanning trees for 3-D cell nuclei segmentation
Conference 2017 10th International Symposium on Image and Signal Processing and Analysis (ISPA), Sep 2017, Ljubljana, France. pp.195-199
@inproceedings{abreu:hal-05493173, TITLE = {{Optimal cut in minimum spanning trees for 3-D cell nuclei segmentation}}, AUTHOR = {Abreu, A. and Frenois, F.-X. and Valitutti, S. and Brousset, P. and Denefle, P. and Naegel, B. and Wemmert, C.}, URL = {https://univoak.hal.science/hal-05493173}, BOOKTITLE = {{2017 10th International Symposium on Image and Signal Processing and Analysis (ISPA)}}, ADDRESS = {Ljubljana, France}, PUBLISHER = {{IEEE}}, PAGES = {195-199}, YEAR = {2017}, MONTH = Sep, DOI = {10.1109/ISPA.2017.8073594}, HAL_ID = {hal-05493173}, HAL_VERSION = {v1}, } -
Journal Breast Cancer Research and Treatment, 2017, 164 (2), pp.305 - 315
Purpose: To improve microscopic evaluation of immune cells relevant in breast cancer oncoimmunology, we aim at distinguishing normal infiltration patterns from lymphocytic lobulitis by advanced image analysis. We consider potential immune cell variations due to the menstrual cycle and oral contraceptives in non-neoplastic mammary gland tissue. Methods: Lymphocyte and macrophage distributions were analyzed in the anatomical context of the resting mammary gland in immunohistochemically stained digital whole slide images obtained from 53 reduction mammoplasty specimens. Our image analysis workflow NS Schaadt · A Grote · N Krönke · HH Kreipe · included automated regions of interest detection, immune cell recognition, and co-registration of regions of interest. Results: In normal lobular epithelium, seven CD8 + lymphocytes per 100 epithelial cells were present on average and about 70% of this T-lymphocyte population was lined up along the basal cell layer in close proximity to the epithelium. The density of CD8 + T-cell was 1.6 fold higher in the luteal than in the follicular phase in spontaneous menstrual cycles and 1.4 fold increased under influence of oral contraceptives, and not co-localized with epithelial proliferation. CD4 + T-cells were infrequent. Abundant CD163 + macrophages were widely spread, including the interstitial compartment, with minor variation during the menstrual cycle. Conclusions: Spatial patterns of different immune cell subtypes determine the range of normal, as opposed to inflammatory conditions of the breast tissue microen-vironment. Advanced image analysis enables quantifi-cation of hormonal effects, refines lymphocytic lobulitis, and shows potential for comprehensive biopsy evaluation in oncoimmunology. Keywords lymphocytic lobulitis · digital pathology · oncoimmunology · object-based image analysis · menstrual cycle · hormonal fluctuations This is the author's version of an article published in Breast Cancer Research and Treatment. The final authenticated version is available online at: https://doi.
@article{schaadt:hal-01875860, TITLE = {{Image analysis of immune cell patterns in the human mammary gland during the menstrual cycle refines lymphocytic lobulitis}}, AUTHOR = {Schaadt, Nadine S. and Alfonso, Juan Carlos L{\'o}pez and Sch{\"o}nmeyer, Ralf and Grote, Anne and Forestier, Germain and Wemmert, C{\'e}dric and Kr{\"o}nke, Nicole and Stoeckelhuber, Mechthild and Kreipe, Hans and Hatzikirou, Haralampos and Feuerhake, Friedrich}, URL = {https://hal.science/hal-01875860}, JOURNAL = {{Breast Cancer Research and Treatment}}, PUBLISHER = {{Springer Verlag}}, VOLUME = {164}, NUMBER = {2}, PAGES = {305 - 315}, YEAR = {2017}, MONTH = Jul, DOI = {10.1007/s10549-017-4239-z}, PDF = {https://hal.science/hal-01875860v1/file/bcrt2017.pdf}, HAL_ID = {hal-01875860}, HAL_VERSION = {v1}, }
2016
-
Journal Scientific Reports, 2016, 6 (1)
Scattered inflammatory cells are commonly observed in mammary gland tissue, most likely in response to normal cell turnover by proliferation and apoptosis, or as part of immunosurveillance. In contrast, lymphocytic lobulitis (LLO) is a recurrent inflammation pattern, characterized by lymphoid cells infiltrating lobular structures, that has been associated with increased familial breast cancer risk and immune responses to clinically manifest cancer. The mechanisms and pathogenic implications related to the inflammatory microenvironment in breast tissue are still poorly understood. Currently, the definition of inflammation is mainly descriptive, not allowing a clear distinction of LLO from physiological immunological responses and its role in oncogenesis remains unclear. To gain insights into the prognostic potential of inflammation, we developed an agent-based model of immune and epithelial cell interactions in breast lobular epithelium. Physiological parameters were calibrated from breast tissue samples of women who underwent reduction mammoplasty due to orthopedic or cosmetic reasons. The model allowed to investigate the impact of menstrual cycle length and hormone status on inflammatory responses to cell turnover in the breast tissue. Our findings suggested that the immunological context, defined by the immune cell density, functional orientation and spatial distribution, contains prognostic information previously not captured by conventional diagnostic approaches. Several studies provided conclusive evidence that a delicate balance between mammary epithelial cell proliferation and apoptosis regulates homeostasis in the healthy breast tissue 1-7. After menarche, and in the absence of pregnancy, the adult female mammary gland is subjected to cyclic fluctuations depending on hormonal stimulation 1,8. In response to such systemic hormonal changes, the breast epithelium undergoes a tightly regulated sequence of cell proliferation and apoptosis during each ovarian/menstrual cycle 1-3. The peak of epithelial cell proliferation has been reported to occur during the luteal phase, suggesting a synergistic influence of steroid hormones, such as estrogen and progesterone 2-5. In turn, the peak of apoptotic activity would be expected in response to decreasing hormone levels towards the end of the menstrual cycle 2-5. However, recent histologic findings indicate that apoptosis reaches its maximum levels in the middle of the luteal phase, although there is also a peak at about the third day of the menstrual cycle 6,7. Experimental measurements of cell turnover, i.e. programmed cell death and proliferation, demonstrated that an imbalance between the mitotic and apoptotic activity might lead to malignant transformation of epithelial cells and tumorigenic processes 9-11. Indeed, excessive cell proliferation promotes accumulation of DNA damage due to insufficient timely repair and mutations 12,13. There is also recent evidence that hormones suppress effective DNA repair and alter DNA damage response (DDR) 13-15 .
@article{alfonso:hal-01878620, TITLE = {{In-silico insights on the prognostic potential of immune cell infiltration patterns in the breast lobular epithelium}}, AUTHOR = {Alfonso, Juan Carlos L{\'o}pez and Schaadt, Nadine S and Sch{\"o}nmeyer, Ralf and Brieu, Nicolas and Forestier, Germain and Wemmert, C{\'e}dric and Feuerhake, Friedrich and Hatzikirou, Haralampos H}, URL = {https://hal.science/hal-01878620}, JOURNAL = {{Scientific Reports}}, PUBLISHER = {{Nature Publishing Group}}, VOLUME = {6}, NUMBER = {1}, YEAR = {2016}, MONTH = Dec, DOI = {10.1038/srep33322}, PDF = {https://hal.science/hal-01878620v1/file/srep2016.pdf}, HAL_ID = {hal-01878620}, HAL_VERSION = {v1}, } -
Detection of lobular structures in normal breast tissue
Journal Computers in Biology and Medicine, 2016, 74, pp.91-102
Background: Ongoing research into inflammatory conditions raises an increasing need to evaluate immune cells in histological sections in biologically relevant regions of interest (ROIs). Herein, we compare different approaches to automatically detect lobular structures in human normal breast tissue in digitized whole slide images (WSIs). This automation is required to perform objective and consistent quantitative studies on large data sets.Methods: In normal breast tissue from nine healthy patients immunohis-tochemically stained for different markers, we evaluated and compared three different image analysis methods to automatically detect lobular structures in WSIs: (1) a bottom–up approach using the cell–based data for subsequent tissue level classification, (2) a top–down method starting with texture classification at tissue level analysis of cell densities in specific ROIs, and (3) a direct texture classification using deep learning technology.Results: All three methods result in comparable overall quality allowing automated detection of lobular structures with minor advantage in sensitivity (approach 3), specificity (approach 2), or processing time (approach 1). Combining the outputs of the approaches further improved the precision. Conclusions: Different approaches of automated ROI detection are feasible and should be selected according to the individual needs of biomarker research. Additionally, detected ROIs could be used as a basis for quantification of immune infiltration in lobular structures.
@article{apou:hal-01321053, TITLE = {{Detection of lobular structures in normal breast tissue}}, AUTHOR = {Apou, Gr{\'e}gory and Schaadt, Nadine S. and Naegel, Beno{\^i}t and Forestier, Germain and Sch{\"o}nmeyer, Ralf and Feuerhake, Friedrich and Wemmert, C{\'e}dric and Grote, Anne}, URL = {https://hal.science/hal-01321053}, JOURNAL = {{Computers in Biology and Medicine}}, PUBLISHER = {{Elsevier}}, VOLUME = {74}, PAGES = {91-102}, YEAR = {2016}, DOI = {10.1016/j.compbiomed.2016.05.004}, PDF = {https://hal.science/hal-01321053v1/file/CBM-2016-HAL.pdf}, HAL_ID = {hal-01321053}, HAL_VERSION = {v1}, } -
Semi-supervised learning using multiple clusterings with limited labeled data
Journal Information Sciences, 2016, 361–362, pp.48-65
Supervised classification consists in learning a predictive model using a set of labeled samples. It is accepted that predictive models accuracy usually increases as more labeled samples are available. Labelled samples are generally difficult to obtain as the labelling step if often performed manually. On the contrary, unlabeled samples are easily available. As the labeling task is tedious and time consuming, users generally provide a very limited number of labeled objects. However, designing approaches able to work efficiently with a very limited number of labeled samples is highly challenging. In this context, semi-supervised approaches have been proposed to leverage from both labeled and unlabeled data. In this paper, we focus on cases where the number of labeled samples is very limited. We review and formalize eight semi-supervised learning algorithms and introduce a new method that combine supervised and unsupervised learning in order to use both labeled and unlabeled data. The main idea of this method is to produce new features derived from a first step of data clustering. These features are then used to enrich the description of the input data leading to a better use of the data distribution. The efficiency of all the methods is compared on various artificial, UCI datasets, and on the classification of a very high resolution remote sensing image. The experiments reveal that our method shows good results, especially when the number of labeled sample is very limited. It also confirms that combining labeled and unlabeled data is very useful in pattern recognition.
@article{forestier:hal-01355189, TITLE = {{Semi-supervised learning using multiple clusterings with limited labeled data}}, AUTHOR = {Forestier, Germain and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-01355189}, JOURNAL = {{Information Sciences}}, PUBLISHER = {{Elsevier}}, VOLUME = {361--362}, PAGES = {48--65}, YEAR = {2016}, DOI = {10.1016/j.ins.2016.04.040}, KEYWORDS = {semi-supervised learning ; classification ; pattern recognition ; remote sensing}, PDF = {https://hal.science/hal-01355189v2/file/is2016.pdf}, HAL_ID = {hal-01355189}, HAL_VERSION = {v2}, }
2015
-
Synthesizing whole slide images
Conference International Symposium on Image and Signal Processing and Analysis (ISPA), Sep 2015, Zagreb, France. pp.154-159
The increasing availability of digital whole slide images opens new perspectives for computer-assisted image analysis complementing modern histopathology, assuming we can implement reliable and efficient image analysis algorithms to extract the biologically relevant information. Both validation and supervised learning techniques typically rely on ground truths manually made by human experts. However, this task is difficult, subjective and usually not exhaustive. This is a well-known issue in the field of biomedical imaging, and a common solution is the use of artificial "phantoms". Following this trend, we study the feasibility of synthesizing artificial histological images to create perfect ground truths. In this paper, we show that it is possible to generate a synthetic whole slide image with reasonable computing resources, and we propose a way to evaluate its quality.
@inproceedings{apou:hal-03811521, TITLE = {{Synthesizing whole slide images}}, AUTHOR = {Apou, Gregory and Feuerhake, Friedrich and Forestier, Germain and Naegel, Benoit and Wemmert, Cedric}, URL = {https://hal.science/hal-03811521}, BOOKTITLE = {{International Symposium on Image and Signal Processing and Analysis (ISPA)}}, ADDRESS = {Zagreb, France}, PUBLISHER = {{IEEE}}, PAGES = {154-159}, YEAR = {2015}, MONTH = Sep, DOI = {10.1109/ispa.2015.7306050}, PDF = {https://hal.science/hal-03811521v1/file/ispa2015.pdf}, HAL_ID = {hal-03811521}, HAL_VERSION = {v1}, } -
A Layered Architecture for a Fuzzy Semantic Approach for Satellite Image Analysis
Journal International Journal of Knowledge and Systems Science, 2015, pp.31-56
The extended use of high and very high spatial resolution imagery inherently demands the adoption of classification methods capable of capturing the underlying semantic. Object-oriented classification methods are currently considered as the most appropriate alternative, due to the incorporation of contextual information and domain knowledge into the analysis. Integrating knowledge initially requires a detailed process of acquisition and later the achievement of a formal representation. Ontologies constitute a very suitable approach to address both knowledge formalization and exploitation. A novel semi-automatic fuzzy semantic approach focused on the extraction and classification of urban objects is hereby introduced. The use of a four-layered architecture allows the separation of concerns among knowledge, rules, experience and meta-knowledge. Knowledge represents the fundamental layer with which the other layers interact. Rules are meant to derive conclusions and make assertions based on knowledge. The experience layer supports the classification process in case of failure when attempting to identify an object, by applying specific expert rules to infer unusual membership. Finally, the meta-knowledge layer contains knowledge about the use of the other layers.
@article{zannimerk:hal-04608916, TITLE = {{A Layered Architecture for a Fuzzy Semantic Approach for Satellite Image Analysis}}, AUTHOR = {Zanni-Merk, Cecilia and Marc-Zwecker, Stella and Wemmert, C{\'e}dric and de Bertrand de Beuvron, Fran{\c c}ois}, URL = {https://hal.science/hal-04608916}, JOURNAL = {{International Journal of Knowledge and Systems Science}}, PAGES = {31-56}, YEAR = {2015}, DOI = {10.4018/IJKSS.2015040103}, HAL_ID = {hal-04608916}, HAL_VERSION = {v1}, } -
Journal Revue des Nouvelles Technologies de l'Information, 2015, RNTI-E27, pp.23-42 in French
Les nouveaux capteurs satellitaires permettent l'acquisition d'images d'un très haut niveau de détail à des cadences élevés, produisant ainsi une importante masse de données. Le traitement manuel de ces données étant devenu impossible, de nouveaux outils sont nécessaires afin de les traiter automatiquement. Dans ce cadre, des algorithmes de segmentation efficaces sont nécessaires pour extraire des objets d'intérêt de ces images. Cependant, les segments produits ne correspondent généralement pas aux objets d'intérêt. Dans cet article, nous proposons de changer le niveau d'abstraction afin d'interpréter les objets d'intérêt comme des objets composés par des segments. Pour cela, nous avons mis en place un processus d'apprentissage multi-niveaux, basé sur des connaissances expertes, dans le but d'apprendre des règles de compositions définissant des objets d'intérêt. Pour gérer l'imprécision relative à l'analyse d'images de télédétection nous proposons d'utiliser la logique floue afin de modéliser les règles de composition. La méthode proposée est validée sur des données de synthèse ainsi que sur des données réelles.
@article{belarte:hal-05417419, TITLE = {{Apprentissage de r{\`e}gles floues pour caract{\'e}riser des objets d'int{\'e}r{\^e}t dans une image de t{\'e}l{\'e}d{\'e}tection}}, AUTHOR = {Belarte, Bruno and Wemmert, C{\'e}dric and Forestier, Germain and Weber, Christiane and Grizonnet, Manuel}, URL = {https://hal.science/hal-05417419}, JOURNAL = {{Revue des Nouvelles Technologies de l'Information}}, PUBLISHER = {{Editions RNTI}}, VOLUME = {RNTI-E27}, PAGES = {23-42}, YEAR = {2015}, HAL_ID = {hal-05417419}, HAL_VERSION = {v1}, }
2014
-
A Fuzzy-Rule Based Ontology for Urban Object Recognition
Conference KEOD 2014 (Knowledge Engineering and Ontology Devlopment), Oct 2014, Rome, Italy. pp.153-160
In this paper we outline the principles of a methodology for semi-automatic recognition of urban objects from satellite images. The methodology aims to provide a framework for bridging the semantic gap problem. Its principle consists in linking abstract geographical domain concepts with image segments, by the means of ontologies use. The imprecision of image data and of qualitative rules formulated by experts geographers are handled by fuzzy logic mechanisms. We have defined fuzzy rules, implemented in SWRL (Semantic Web Rule Language), which allow classification of image segments in the ontology. We propose some fuzzy classification strategies, which are compared and evaluated through an experimentation performed on a VHR image of Strasbourg region.
@inproceedings{marczwecker:hal-05403437, TITLE = {{A Fuzzy-Rule Based Ontology for Urban Object Recognition}}, AUTHOR = {Marc-Zwecker, Stella and Asnoune, Khalid and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-05403437}, BOOKTITLE = {{IC3K 2014: Proceedings of the International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management - Volume 2}}, ADDRESS = {Rome, Italy}, PUBLISHER = {{SCITEPRESS}}, PAGES = {153-160}, YEAR = {2014}, MONTH = Oct, DOI = {10.5220/0005026601530160}, HAL_ID = {hal-05403437}, HAL_VERSION = {v1}, } -
Efficient Region-based Classification for Whole Slide Images
Conference International Joint Conference on Computer Vision, Imaging and Computer Graphics, 5-8 january 2017, Lisbon, Portugal, Jan 2014, Lisbonne, Portugal. pp.239-256
@inproceedings{apou:hal-03797892, TITLE = {{Efficient Region-based Classification for Whole Slide Images}}, AUTHOR = {Apou, Gr{\'e}gory and Naegel, Beno{\^i}t and Forestier, Germain and Feuerhake, Friedrich and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-03797892}, BOOKTITLE = {{Computer Vision, Imaging and Computer Graphics - Theory and Applications}}, ADDRESS = {Lisbonne, Portugal}, PUBLISHER = {{Springer International Publishing}}, SERIES = {Communications in Computer and Information Science}, VOLUME = {550}, PAGES = {239-256}, YEAR = {2014}, MONTH = Jan, DOI = {10.1007/978-3-319-25117-2\_15}, HAL_ID = {hal-03797892}, HAL_VERSION = {v1}, } -
Fast segmentation for texture-based cartography of whole slide images
Conference International Conference on Computer Vision Theory and Applications (VISAPP), Jan 2014, Lisbon, Portugal. pp.309-319
In recent years, new optical microscopes have been developed, providing very high spatial resolution images called Whole Slide Images (WSI). The fast and accurate display of such images for visual analysis by pathologists and the conventional automated analysis remain challenging, mainly due to the image size (sometimes billions of pixels) and the need to analyze certain image features at high resolution. To propose a decision support tool to help the pathologist interpret the information contained by the WSI, we present a new approach to establish an automatic cartography of WSI in reasonable time. The method is based on an original segmentation algorithm and on a supervised multiclass classification using a textural characterization of the regions computed by the segmentation. Application to breast cancer WSI shows promising results in terms of speed and quality.
@inproceedings{apou:hal-04524187, TITLE = {{Fast segmentation for texture-based cartography of whole slide images}}, AUTHOR = {Apou, Gr{\'e}gory and Naegel, Benoit and Forestier, Germain and Feuerhake, Friedrich and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-04524187}, BOOKTITLE = {{International Conference on Computer Vision Theory and Applications (VISAPP)}}, ADDRESS = {Lisbon, Portugal}, PUBLISHER = {{IEEE Xplore}}, PAGES = {309-319}, YEAR = {2014}, MONTH = Jan, DOI = {10.5220/0004687403090319}, KEYWORDS = {Classification ; Whole Slide Images ; Biomedical Image Processing ; Segmentation}, PDF = {https://hal.science/hal-04524187v1/file/islandora_59843.pdf}, HAL_ID = {hal-04524187}, HAL_VERSION = {v1}, } -
Template Based Hierarchical Building Extraction
Journal IEEE Geoscience and Remote Sensing Letters, 2014, 11 (3), pp.706--710
@article{sellaouti:hal-05404692, TITLE = {{Template Based Hierarchical Building Extraction}}, AUTHOR = {Sellaouti, A. and Hamouda, A. and Deruyver, A. and Wemmert, C.}, URL = {https://hal.science/hal-05404692}, JOURNAL = {{IEEE Geoscience and Remote Sensing Letters}}, PUBLISHER = {{IEEE - Institute of Electrical and Electronics Engineers}}, VOLUME = {11}, NUMBER = {3}, PAGES = {706--710}, YEAR = {2014}, DOI = {10.1109/LGRS.2013.2276936}, HAL_ID = {hal-05404692}, HAL_VERSION = {v1}, }
2013
-
Stain unmixing in brightfield multiplexed immunohistochemistry
Conference IEEE International Conference on Image Processing (ICIP), Sep 2013, Melbourne, Australia. pp.1125-1129
Automated image analysis of multiplexed brightfield immunohistochemistry assays is a challenging objective. One central task of the analysis is the robust identification of the different stains in the image, called stain unmixing. Stain unmixing strongly depends on the method of image acquisition. Currently available multispectral cameras enable color unmixing of single fields of view (FoV), selected by matter experts (e.g. pathologists). Beyond the individual FoV approach, there is an increasing need to process larger regions or whole histopathological sections (whole slide imaging; WSI). Rapid color deconvolution in WSI is a challenge that is only partially solved. We propose a method based on a multilayer perceptron to compute dye-specific stain layers for chromogenic red and brown labeling in WSI.
@inproceedings{wemmert:hal-03825906, TITLE = {{Stain unmixing in brightfield multiplexed immunohistochemistry}}, AUTHOR = {Wemmert, Cedric and Kruger, Juliane M and Forestier, Germain and Sternberger, Ludovic and Feuerhake, Friedrich and Gancarski, Pierre}, URL = {https://hal.science/hal-03825906}, BOOKTITLE = {{IEEE International Conference on Image Processing (ICIP)}}, ADDRESS = {Melbourne, Australia}, PUBLISHER = {{IEEE}}, PAGES = {1125-1129}, YEAR = {2013}, MONTH = Sep, DOI = {10.1109/icip.2013.6738232}, KEYWORDS = {multilayer perceptron ; color deconvolution ; immunohistochemistry ; biomedical imaging}, PDF = {https://hal.science/hal-03825906v1/file/icip2013.pdf}, HAL_ID = {hal-03825906}, HAL_VERSION = {v1}, } -
Learning fuzzy rules to characterize objects of interest from remote sensing images
Conference IEEE International Geoscience and Remote Sensing Symposium (IGARSS), Jul 2013, Melbourne, France
In this article a new method for learning concepts from examples of objects provided by experts for remote sensing images is presented. The goal of this method is to give the geographer expert a description of complex objects of interest extracted from very high resolution remote sensing images. The description of such objects needs to handle imprecision inherent to segmentation and very high resolution images. The first step of this approach is to classify objects composing all the examples. This classification allows the learning of a rule describing how the examples are composed regarding the segmentation. Finally, this rule is used to extract objects corresponding to the examples. Experiments on a remote sensing image of a urban landscape in Toulouse, France are presented to show the relevance of the method.
@inproceedings{belarte:hal-01888500, TITLE = {{Learning fuzzy rules to characterize objects of interest from remote sensing images}}, AUTHOR = {Belarte, B. and Wemmert, C. and Forestier, G. and Grizonnet, M. and Weber, C.}, URL = {https://hal.science/hal-01888500}, BOOKTITLE = {{2013 IEEE International Geoscience and Remote Sensing Symposium - IGARSS}}, ADDRESS = {Melbourne, France}, PUBLISHER = {{IEEE}}, YEAR = {2013}, MONTH = Jul, DOI = {10.1109/IGARSS.2013.6723453}, PDF = {https://hal.science/hal-01888500v1/file/igarss2013.pdf}, HAL_ID = {hal-01888500}, HAL_VERSION = {v1}, } -
Comparison of optical sensors discrimination ability using spectral libraries
Journal International Journal of Remote Sensing, 2013, 34 (7), pp.2327 - 2349
In remote sensing, the ability to discriminate different land covers or material types is directly linked with the spectral resolution and sampling provided by the optical sensor. Previous studies showed that the spectral resolution is a critical issue, especially in complex environment. In spite of the increasing availability of hyperspectral data, multispectral optical sensors onboard various satellites are acquiring everyday a massive amount of data with a relatively poor spectral resolution (i.e. usually about 4 to 7 spectral bands). These remotely sensed data are intensively used for Earth observation regardless of their limited spectral resolution. In this paper, we studied seven of these optical sensors: Pleiades, QuickBird, SPOT5, Ikonos, Landsat TM, Formosat and Meris. This study focuses on the ability of each sensor to discriminate different materials according to its spectral resolution. We used four different spectral libraries which contains around 2500 spectra of materials and land covers with a fine spectral resolution. These spectra were convolved with the Relative Spectral Responses (RSR) of each sensor to create spectra at the sensors' resolutions. Then, these reduced spectra were compared using separability indexes (Divergence, Transformed divergence, Bhattacharyya, Jeffreys-Matusita) and machine learning tools. In the experiments, we highlighted that the spectral bands configuration could lead to important differences in classification accuracy according to the context of application (e.g. urban area).
@article{forestier:hal-01875856, TITLE = {{Comparison of optical sensors discrimination ability using spectral libraries}}, AUTHOR = {Forestier, Germain and Inglada, Jordi and Wemmert, C{\'e}dric and Gancarski, Pierre}, URL = {https://hal.science/hal-01875856}, JOURNAL = {{International Journal of Remote Sensing}}, PUBLISHER = {{Taylor \& Francis}}, VOLUME = {34}, NUMBER = {7}, PAGES = {2327 - 2349}, YEAR = {2013}, MONTH = Apr, DOI = {10.1080/01431161.2012.744488}, PDF = {https://hal.science/hal-01875856v1/file/ijrs2013.pdf}, HAL_ID = {hal-01875856}, HAL_VERSION = {v1}, } -
Coastal image interpretation using background knowledge and semantics
Journal Computers & Geosciences, 2013, 54, pp.88 - 96
In this article, we present a framework to model and to use knowledge provided by experts for remote sensing image interpretation of coastal area. The goal of this approach is to associate semantic to regions issued from the segmentation of an image. The idea is to start with a raw description of the knowledge given by the expert on the different thematic object classes present in the image. This knowledge is then decomposed and formalized to be usable during the classification process. A first interpretation of the image is computed through an ontology with spectral information about the classes. Then, a set of Knowledge Functions (KFs) are defined according to the description of the expert's knowledge. These KFs are then used to check the consistency of the spectral interpretation and to detect potentially mislabeled regions. The interpretation of these regions is revised in an iterative process to produce a more accurate final result. Experiments on remote sensing images of a coastal zone of Normandy, France are presented to show the relevance of the method.
@article{forestier:hal-01875858, TITLE = {{Coastal image interpretation using background knowledge and semantics}}, AUTHOR = {Forestier, Germain and Wemmert, C{\'e}dric and Puissant, Anne}, URL = {https://hal.science/hal-01875858}, JOURNAL = {{Computers \& Geosciences}}, PUBLISHER = {{Elsevier}}, VOLUME = {54}, PAGES = {88 - 96}, YEAR = {2013}, MONTH = Apr, DOI = {10.1016/j.cageo.2012.11.023}, KEYWORDS = {knowledge-based image interpretation ; ontology ; coastal management}, PDF = {https://hal.science/hal-01875858v1/file/cageo2013.pdf}, HAL_ID = {hal-01875858}, HAL_VERSION = {v1}, } -
Journal The Journal of Pathology, 2013, 229 (4), pp.569-578
@article{kruger:hal-03135392, TITLE = {{Combat or surveillance?}}, AUTHOR = {Kr{\"u}ger, Juliane and Wemmert, C{\'e}dric and Sternberger, Ludovic and Bonnas, Christel and Dietmann, Gabriele and Gan{\c c}arski, Pierre and Feuerhake, Friedrich}, URL = {https://hal.science/hal-03135392}, JOURNAL = {{The Journal of Pathology}}, PUBLISHER = {{Wiley}}, VOLUME = {229}, NUMBER = {4}, PAGES = {569-578}, YEAR = {2013}, MONTH = Feb, DOI = {10.1002/path.4150}, HAL_ID = {hal-03135392}, HAL_VERSION = {v1}, } -
Conference EGC 2013, 2013, Toulouse, France in French
@inproceedings{sellaouti:hal-05394910, TITLE = {{Approche orient{\'e}e objet s{\'e}mantique et coop{\'e}rative pour la classification des images de zones urbaines {\`a} tr{\`e}s haute r{\'e}solution}}, AUTHOR = {Sellaouti, A. and Hamouda, A. and Deruyver, A. and Wemmert, C.}, URL = {https://univoak.hal.science/hal-05394910}, BOOKTITLE = {{EGC 2013}}, ADDRESS = {Toulouse, France}, YEAR = {2013}, HAL_ID = {hal-05394910}, HAL_VERSION = {v1}, } -
Combat or Surveillance? Evaluation of the Heterogeneous Inflammatory Breast Cancer Microenvironment
Journal The Journal of Pathology, 2013, 259 (4), pp.569--578
@article{kruger:hal-05398236, TITLE = {{Combat or Surveillance? Evaluation of the Heterogeneous Inflammatory Breast Cancer Microenvironment}}, AUTHOR = {Kr{\"u}ger, J. and Wemmert, C. and Sternberger, L. and Bonnas, C. and Dietmann, G. and Gan{\c c}arski, P. and Feuerhake, F.}, URL = {https://univoak.hal.science/hal-05398236}, JOURNAL = {{The Journal of Pathology}}, PUBLISHER = {{Wiley}}, VOLUME = {259}, NUMBER = {4}, PAGES = {569--578}, YEAR = {2013}, HAL_ID = {hal-05398236}, HAL_VERSION = {v1}, } -
Towards a semi-automatic semantic approach to satellite images analysis
Conference KES 2013 - 17th International Conference in Knowledge Based and Intelligent Information and Engineering Sy, 2013, Kitakyushu, Japan. pp.1388--1397
@inproceedings{disciascio:hal-05395113, TITLE = {{Towards a semi-automatic semantic approach to satellite images analysis}}, AUTHOR = {Di Sciascio, M. and Zanni-Merk, C. and Wemmert, C. and Marc-Zwecker, S. and de Bertrand de Beuvron, F.}, URL = {https://univoak.hal.science/hal-05395113}, BOOKTITLE = {{KES 2013 - 17th International Conference in Knowledge Based and Intelligent Information and Engineering Sy}}, ADDRESS = {Kitakyushu, Japan}, PUBLISHER = {{Procedia Computer Science}}, VOLUME = {22}, PAGES = {1388--1397}, YEAR = {2013}, DOI = {10.1016/j.procs.2013.11.057}, HAL_ID = {hal-05395113}, HAL_VERSION = {v1}, }
2012
-
Knowledge-based region labeling for remote sensing image interpretation
Journal Computers, Environment and Urban Systems, 2012, 36 (5), pp.470 - 480
The increasing availability of High Spatial Resolution (HSR) satellite images is an opportunity to characterize and identify urban objects. Thus, the augmentation of the precision led to a need of new image analysis methods using region-based (or object-based) approaches. In this field, an important challenge is the use of domain knowledge for automatic urban objects identification, and a major issue is the formalization and exploitation of this knowledge. In this paper, we present the building steps of a knowledge-base of urban objects allowing to perform the interpretation of HSR images in order to help urban planners to automatically map the territory. The knowledge-base is used to assign segmented regions (i.e. extracted from the images) into semantic objects (i.e. concepts of the knowledge-base). A matching process between the regions and the concepts of the knowledge-base is proposed, allowing to bridge the semantic gap between the images content and the interpretation. The method is validated on Quickbird images of the urban areas of Strasbourg and Marseille (France). The results highlight the capacity of the method to automatically identify urban objects using the domain knowledge. This is the author's version of an article published in Computers, Environment and Urban Systems. The final authenticated version is available online at: http://dx.
@article{forestier:hal-01875854, TITLE = {{Knowledge-based region labeling for remote sensing image interpretation}}, AUTHOR = {Forestier, Germain and Puissant, Anne and Wemmert, C{\'e}dric and Gan{\c c}arski, Pierre}, URL = {https://hal.science/hal-01875854}, JOURNAL = {{Computers, Environment and Urban Systems}}, PUBLISHER = {{Elsevier}}, VOLUME = {36}, NUMBER = {5}, PAGES = {470 - 480}, YEAR = {2012}, MONTH = Sep, DOI = {10.1016/j.compenvurbsys.2012.01.003}, KEYWORDS = {region labeling ; semantic interpretation ; remote sensing images ; high resolution ; knowledge base ; Urban object}, PDF = {https://hal.science/hal-01875854v1/file/ceus2012.pdf}, HAL_ID = {hal-01875854}, HAL_VERSION = {v1}, }
2011
-
8ème édition de l'atelier "Fouille de données complexes" : complexité liée aux données multiples
Other 2011, 146p in French
L'atelier sur la fouille de données complexes est proposé à l'instigation du groupe de travail EGC "Fouille de données complexes". Chaque année les organisateurs proposent une thématique de recherche qui suscite l'intérêt des chercheurs et des industriels. La précédente édition de l'atelier ayant permis de fédérer avec succès des recherches portant sur la problématique de complexité liée à la multiplicité des données, nous avons choisit de renouveler l'affichage de cette priorité thématique pour cette huitième édition.
@misc{cleuziou:hal-00609725, TITLE = {{8{\`e}me {\'e}dition de l'atelier ''Fouille de donn{\'e}es complexes'' : complexit{\'e} li{\'e}e aux donn{\'e}es multiples}}, AUTHOR = {Cleuziou, Guillaume and de Runz, Cyril and Lebbah, Mustapha and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-00609725}, NOTE = {Actes du 8{\`e}me atelier Fouille de donn{\'e}es complexes - complexit{\'e} li{\'e}e aux donn{\'e}es multiples - dans le cadre de la conf{\'e}rence EGC'2011.}, PAGES = {146p}, YEAR = {2011}, MONTH = Jan, PDF = {https://hal.science/hal-00609725v1/file/actes.pdf}, HAL_ID = {hal-00609725}, HAL_VERSION = {v1}, }
2010
-
Background Knowledge Integration in Clustering Using Purity Indexes
Conference International Conference on Knowledge Science, Engineering and Management (KSEM), Sep 2010, Belfast, United Kingdom. pp.28 - 38
In recent years, the use of background knowledge to improve the data mining process has been intensively studied. Indeed, background knowledge along with knowledge directly or indirectly provided by the user are often available. However, it is often difficult to formalize this kind of knowledge, as it is often dependent of the domain. In this article, we studied the integration of knowledge as labeled objects in clustering algorithms. Several criteria allowing the evaluation of the purity of a clustering are presented and their behaviours are compared using artificial datasets. Advantages and drawbacks of each criterion are analyzed in order to help the user to make a choice among them.
@inproceedings{forestier:hal-03813327, TITLE = {{Background Knowledge Integration in Clustering Using Purity Indexes}}, AUTHOR = {Forestier, Germain and Wemmert, C{\'e}dric and Gan{\c c}arski, Pierre}, URL = {https://hal.science/hal-03813327}, BOOKTITLE = {{International Conference on Knowledge Science, Engineering and Management (KSEM)}}, ADDRESS = {Belfast, United Kingdom}, PUBLISHER = {{Springer Berlin Heidelberg}}, SERIES = {Lecture Notes in Computer Science}, VOLUME = {6291}, PAGES = {28 - 38}, YEAR = {2010}, MONTH = Sep, DOI = {10.1007/978-3-642-15280-1\_6}, KEYWORDS = {Clustering ; background knowledge ; semi-supervised algorithm ; purity indexes}, PDF = {https://hal.science/hal-03813327v1/file/ksem2010.pdf}, HAL_ID = {hal-03813327}, HAL_VERSION = {v1}, } -
Towards conflict resolution in collaborative clustering
Conference IEEE International Conference Intelligent Systems (IS), Jul 2010, London, United Kingdom. pp.361-366
In recent years, a lot of work has focused on the use of multiple clusterings for partitioning data. These approaches are supported by the existence of a huge number of clustering algorithms. Thus, different methods have been proposed to create alternative clustering results from the same data. However, the different clustering results are usually generated without sharing information and the user is often asked to select the final result. To cope with these issues, a new paradigm named collaborative clustering has been proposed recently. In collaborative clustering, different clustering methods work together (i.e. collaborate) to reach an agreement on the clustering of a common dataset. At the end of the collaboration, the results are expected to be strongly similar. In this paper, we address the problem of the collaboration of different clustering methods and we compare four collaboration strategies. Our experiments compare the different strategies on synthetic and real-life datasets and provide insight into the advantages and the drawbacks of each strategy.
@inproceedings{forestier:hal-03813345, TITLE = {{Towards conflict resolution in collaborative clustering}}, AUTHOR = {Forestier, Germain and Wemmert, Cedric and Gancarski, Pierre}, URL = {https://hal.science/hal-03813345}, BOOKTITLE = {{IEEE International Conference Intelligent Systems (IS)}}, ADDRESS = {London, United Kingdom}, PUBLISHER = {{IEEE}}, PAGES = {361-366}, YEAR = {2010}, MONTH = Jul, DOI = {10.1109/is.2010.5548343}, PDF = {https://hal.science/hal-03813345v1/file/is2010.pdf}, HAL_ID = {hal-03813345}, HAL_VERSION = {v1}, } -
Collaborative clustering with background knowledge
Journal Data and Knowledge Engineering, 2010, 69 (2), pp.211 - 228
The aim of collaborative clustering is to make different clustering methods collaborate, in order to reach at an agreement on the partitioning of a common dataset. As different clustering methods can produce different partitioning of the same dataset, finding a consensual clustering from these results is often a hard task. The collaboration aims to make the methods agree on the partitioning through a refinement of their results. This process tends to make the results more similar. In this paper, after the introduction of the collaboration process, we present different ways to integrate background knowledge into it. Indeed, in recent years, the integration of background knowledge in clustering algorithms has been the subject of a lot of interest. This integration often leads to an improvement of the quality of the results. We discuss how such integration in the collaborative process is beneficial and we present experiments in which background knowledge is used to guide collaboration.
@article{forestier:hal-01875853, TITLE = {{Collaborative clustering with background knowledge}}, AUTHOR = {Forestier, Germain and Gan{\c c}arski, Pierre and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-01875853}, JOURNAL = {{Data and Knowledge Engineering}}, PUBLISHER = {{Elsevier}}, VOLUME = {69}, NUMBER = {2}, PAGES = {211 - 228}, YEAR = {2010}, MONTH = Feb, DOI = {10.1016/j.datak.2009.10.004}, KEYWORDS = {knowledge-guided clustering ; pattern recognition ; classification ; unsupervised learning ; collaborative clustering}, PDF = {https://hal.science/hal-01875853v1/file/dke2010.pdf}, HAL_ID = {hal-01875853}, HAL_VERSION = {v1}, } -
Journal Pattern Recognition Letters, 2010, 31 (15), pp.2364-2374
Automatic image interpretation is often achieved by first performing a segmentation of the image (i.e., gathering neighbouring pixels into homogeneous regions) and then applying a supervised region-based classification. In such a process, the quality of the segmentation step is of great importance in the final classified result. Nevertheless, whereas the classification step takes advantage from some prior knowledge such as learning sample pixels, the segmentation step rarely does. In this paper, we propose to involve such samples through machine learning procedures to improve the segmentation process. More precisely, we consider the watershed transform segmentation algorithm, and rely on both a fuzzy supervised classification procedure and a genetic algorithm in order to respectively build the elevation map used in the watershed paradigm and tune segmentation parameters. We also propose new criteria for segmentation evaluation based on learning samples. We have evaluated our method on remotely sensed images. The results assert the relevance of machine learning as a way to introduce knowledge within the watershed segmentation process.
@article{derivaux:hal-00512749, TITLE = {{Supervised image segmentation using watershed transform, fuzzy classification and evolutionary computation}}, AUTHOR = {Derivaux, S{\'e}bastien and Forestier, Germain and Wemmert, C{\'e}dric and Lef{\`e}vre, S{\'e}bastien}, URL = {https://hal.science/hal-00512749}, JOURNAL = {{Pattern Recognition Letters}}, PUBLISHER = {{Elsevier}}, VOLUME = {31}, NUMBER = {15}, PAGES = {2364-2374}, YEAR = {2010}, DOI = {10.1016/j.patrec.2010.07.007}, KEYWORDS = {supervised image segmentation ; watershed transform ; fuzzy classification ; genetic algorithm}, PDF = {https://hal.science/hal-00512749v2/file/prl2010.pdf}, HAL_ID = {hal-00512749}, HAL_VERSION = {v2}, }
2009
-
Mining spectral libraries to study sensors' discrimination ability
Conference SPIE Europe Remote Sensing, Aug 2009, Berlin, Germany
In remote sensing data classification, the ability to discriminate different land cover or material types is directly linked with the spectral resolution and sampling provided by the optical sensor. 1 Several previous studies 2-4 showed that the spectral resolution is a critical issue, especially to discriminate different land covers in urban areas. In spite of the increasing availability of hyperspectral data, multispectral optical sensors on board of several satellites are still acquiring everyday a massive amount of data with a relatively poor spectral resolution (i.e. usually about 4 to 7 spectral bands). These remotely sensed data are intensively used for Earth observation regardless of their limited spectral resolution. In this paper, we propose to study the capacity of discrimination of several of these optical sensors : Pleiades, QuickBird, SPOT5, Ikonos, Landsat TM, Formosat and Meris. To achieve this goal, we used different spectral libraries which provide spectra of materials and land covers generally with a fine spectral resolution. These spectra were extracted from these libraries and convolved with the Relative Spectral Responses (RSR) of each sensor to create spectra at the sensors' resolutions. Then, these reduced spectra were evaluated thanks to a classical separability index and machine learning tools. This study focuses on the capacity of each sensor to discriminate different materials according to its spectral resolution. As the spectra for each sensor are created from the exact same spectra extracted from the libraries, the only variation is the RSR of the sensors. This approach allows us to fairly compare the different ability of the sensors to discriminate materials.
@inproceedings{forestier:hal-01887503, TITLE = {{Mining spectral libraries to study sensors' discrimination ability}}, AUTHOR = {Forestier, Germain and Inglada, Jordi and Wemmert, C{\'e}dric and Gancarski, Pierre}, URL = {https://hal.science/hal-01887503}, BOOKTITLE = {{SPIE Europe Remote Sensing}}, ADDRESS = {Berlin, Germany}, YEAR = {2009}, MONTH = Aug, DOI = {10.1117/12.830392}, PDF = {https://hal.science/hal-01887503v1/file/spie2009.pdf}, HAL_ID = {hal-01887503}, HAL_VERSION = {v1}, } -
Multiresolution remote sensing image clustering
Journal IEEE Geoscience and Remote Sensing Letters, 2009, 6 (3), pp.533-537
With the multiplication of satellite images with complementary spatial and spectral resolution, a major issue in the classification process is the simultaneous use of several images. In this context, the objective of this letter is to propose a new method which uses information contained in both spatial resolution. The main idea is that on one hand, the semantics level associated with an image depends on its spatial resolution, and on the other hand, information given by these images are complementary. The goal of this multiresolution image method is to automatically build a classification using knowledge extracted from both images, by unsupervised way and without pre-processing image fusion. The method is tested by using a Quickbird (2.8m) and a SPOT-4 (20m)images on the urban area of Strasbourg (France). The experiments have shown that the results are better than a classical unsupervised classification on each image, and comparable to a supervised region-based classification on the HSR image. Index Terms-multiresolution, clustering, remote sensing image analysis, high spatial resolution. This is the author's version of an article published in IEEE Geoscience and Remote Sensing Letters. The final authenticated version is available online at: http://dx.
@article{wemmert:halshs-00520632, TITLE = {{Multiresolution remote sensing image clustering}}, AUTHOR = {Wemmert, C{\'e}dric and Puissant, Anne and Forestier, Germain and Gancarski, Pierre}, URL = {https://shs.hal.science/halshs-00520632}, JOURNAL = {{IEEE Geoscience and Remote Sensing Letters}}, PUBLISHER = {{IEEE - Institute of Electrical and Electronics Engineers}}, VOLUME = {6}, NUMBER = {3}, PAGES = {533-537}, YEAR = {2009}, DOI = {10.1109/LGRS.2009.2020825}, KEYWORDS = {multiresolution ; clustering ; remote sensing image analysis ; high spatial resolution}, PDF = {https://shs.hal.science/halshs-00520632v1/file/grsl2009.pdf}, HAL_ID = {halshs-00520632}, HAL_VERSION = {v1}, }
2008
-
Multisource Images Analysis Using Collaborative Clustering
Journal EURASIP Journal on Advances in Signal Processing, 2008, 2008 (1)
The development of very high-resolution (VHR) satellite imagery has produced a huge amount of data. The multiplication of satellites which embed different types of sensors provides a lot of heterogeneous images. Consequently, the image analyst has often many different images available, representing the same area of the Earth surface. These images can be from different dates, produced by different sensors, or even at different resolutions. The lack of machine learning tools using all these representations in an overall process constraints to a sequential analysis of these various images. In order to use all the information available simultaneously, we propose a framework where different algorithms can use different views of the scene. Each one works on a different remotely sensed image and, thus, produces different and useful information. These algorithms work together in a collaborative way through an automatic and mutual refinement of their results, so that all the results have almost the same number of clusters, which are statistically similar. Finally, a unique result is produced, representing a consensus among the information obtained by each clustering method on its own image. The unified result and the complementarity of the single results (i.e., the agreement between the clustering methods as well as the disagreement) lead to a better understanding of the scene. The experiments carried out on multispectral remote sensing images have shown that this method is efficient to extract relevant information and to improve the scene understanding.
@article{forestier:hal-01878610, TITLE = {{Multisource Images Analysis Using Collaborative Clustering}}, AUTHOR = {Forestier, Germain and Wemmert, C{\'e}dric and Gancarski, Pierre}, URL = {https://hal.science/hal-01878610}, JOURNAL = {{EURASIP Journal on Advances in Signal Processing}}, PUBLISHER = {{SpringerOpen}}, VOLUME = {2008}, NUMBER = {1}, YEAR = {2008}, MONTH = Dec, DOI = {10.1155/2008/374095}, PDF = {https://hal.science/hal-01878610v1/file/eurasip2008.pdf}, HAL_ID = {hal-01878610}, HAL_VERSION = {v1}, } -
On Combining Unsupervised Classification and Ontology Knowledge
Conference IEEE International Geoscience and Remote Sensing Symposium (IGARSS), Jul 2008, Boston, United States
This paper presents a way to combine knowledge obtained from a clustering algorithm and from an ontology. Using the both sources of information allows to improve the results of the knowledge discovery process. The basic property of clustering algorithms, which is to group similar objects, is the key of this approach. We use it to extend the knowledge given by an ontology. Indeed, this knowledge can be partial or not enough accurate, and clustering can then be used to fill this lack of information. We also present results and validation in the field of remote sensing image interpretation.
@inproceedings{forestier:hal-01887351, TITLE = {{On Combining Unsupervised Classification and Ontology Knowledge}}, AUTHOR = {Forestier, Germain and Wemmert, C{\'e}dric and Gancarski, Pierre}, URL = {https://hal.science/hal-01887351}, BOOKTITLE = {{IEEE International Geoscience and Remote Sensing Symposium (IGARSS)}}, ADDRESS = {Boston, United States}, PUBLISHER = {{IEEE}}, YEAR = {2008}, MONTH = Jul, DOI = {10.1109/igarss.2008.4779741}, KEYWORDS = {Knowledge based systems ; Clus-tering methods ; Pattern recognition ; Image classification}, PDF = {https://hal.science/hal-01887351v1/file/igarss2008.pdf}, HAL_ID = {hal-01887351}, HAL_VERSION = {v1}, } -
An Evolutionary Approach for Ontology Driven Image Interpretation
Conference European Workshop on Evolutionary Computation in Image Analysis and Signal Processing, Jun 2008, Naples, Italy. pp.295-305
Image mining and interpretation is a quite complex process. In this article, we propose to model expert knowledge on objects present in an image through an ontology. This ontology will be used to drive a segmentation process by an evolutionary approach. This method uses a genetic algorithm to find segmentation parameters which allow to identify in the image the objects described by the expert in the ontology. The fitness function of the genetic algorithm uses the ontology to evaluate the segmentation. This approach does not need examples and enables to reduce the semantic gap between automatic interpretation of images and expert knowledge.
@inproceedings{forestier:hal-04514827, TITLE = {{An Evolutionary Approach for Ontology Driven Image Interpretation}}, AUTHOR = {Forestier, Germain and Derivaux, S{\'e}bastien and Wemmert, C{\'e}dric and Gan{\c c}arski, Pierre}, URL = {https://hal.science/hal-04514827}, BOOKTITLE = {{Lecture Notes in Computer Science}}, ADDRESS = {Naples, Italy}, PUBLISHER = {{Springer Berlin Heidelberg}}, SERIES = {Lecture Notes in Computer Science}, VOLUME = {4974}, PAGES = {295-305}, YEAR = {2008}, MONTH = Jun, DOI = {10.1007/978-3-540-78761-7\_30}, PDF = {https://hal.science/hal-04514827v1/file/evoiasp2008.pdf}, HAL_ID = {hal-04514827}, HAL_VERSION = {v1}, } -
Extraction de détecteurs d'objets urbains à partir d'une ontologie
Conference Journées Francophones Extraction et Gestion des Connaissances (EGC) Atelier Extraction de COnnaissance à partir d'Images (ECOI), 2008, Sophia Antipolis, France in French
Afin de parvenir à une méthode d'interprétation automatique d'im-ages de télédétection à très haute résolution spatiale, il est nécessaire d'exploiter autant que possible les connaissances du domaine. Pour détecter différents types d'objet comme la route ou le bâti, des méthodes très spécifiques ont été dévelop-pées pour obtenir de très bons résultats. Ces méthodes utilisent des connais-sances du domaine sans les formaliser. Dans cet article, nous proposons tout d'abord de modéliser la connaissance du domaine de manière explicite au sein d'une ontologie. Ensuite, nous introduisons un algorithme pour construire des détecteurs spécifiques utilisant les connaissances de cette ontologie. La sépara-tion nette entre modélisation des connaissances et construction des détecteurs rend plus lisible le processus d'interprétation. Ce découplage permet également d'utiliser l'algorithme de construction de détecteurs dans un autre domaine d'ap-plication, ou de modifier l'algorithme de construction de détecteurs sans modi-fier l'ontologie.
@inproceedings{derivaux:hal-00516099, TITLE = {{Extraction de d{\'e}tecteurs d'objets urbains {\`a} partir d'une ontologie}}, AUTHOR = {Derivaux, S{\'e}bastien and Forestier, Germain and Wemmert, C{\'e}dric and Lef{\`e}vre, S{\'e}bastien}, URL = {https://hal.science/hal-00516099}, BOOKTITLE = {{Journ{\'e}es Francophones Extraction et Gestion des Connaissances (EGC) Atelier Extraction de COnnaissance {\`a} partir d'Images (ECOI),}}, ADDRESS = {Sophia Antipolis, France}, PAGES = {-}, YEAR = {2008}, PDF = {https://hal.science/hal-00516099v1/file/ecoi2008.pdf}, HAL_ID = {hal-00516099}, HAL_VERSION = {v1}, }
2007
-
Ontology-based Object Recognition for Remote Sensing Image Interpretation
Conference IEEE International Conference on Tools with Artificial Intelligence (ICTAI), Oct 2007, Patras, Greece. pp.472-479
The multiplication of Very High Resolution (spatial or spectral) remote sensing images appears to be an opportunity to identify objects in urban and periurban areas. The classification methods applied in the object-oriented image analysis approach could be based on the use of domain knowledge. A major issue in these approaches is domain knowledge formalization and exploitation. In this paper, we propose a recognition method based on an ontology which has been developed by experts of the domain. In order to give objects a semantic meaning, we have developed a matching process between an object and the concepts of the ontology. Experiments are made on a Quickbird image. The quality of the results shows the effectiveness of the proposed method.
@inproceedings{durand:hal-01463170, TITLE = {{Ontology-based Object Recognition for Remote Sensing Image Interpretation}}, AUTHOR = {Durand, Nicolas and Derivaux, S{\'e}bastien and Forestier, Germain and Wemmert, C{\'e}dric and Gancarski, Pierre and Boussa{\"i}d, Omar and Puissant, Anne}, URL = {https://hal.science/hal-01463170}, BOOKTITLE = {{IEEE International Conference on Tools with Artificial Intelligence (ICTAI)}}, ADDRESS = {Patras, Greece}, PUBLISHER = {{IEEE}}, PAGES = {472-479}, YEAR = {2007}, MONTH = Oct, DOI = {10.1109/ICTAI.2007.111}, PDF = {https://hal.science/hal-01463170v2/file/ictai2007.pdf}, HAL_ID = {hal-01463170}, HAL_VERSION = {v2}, } -
Collaborative multi-step mono-level multi-strategy classification
Journal Multimedia Tools and Applications, 2007, 35 (1), pp.1-27
@article{gancarski:hal-03135400, TITLE = {{Collaborative multi-step mono-level multi-strategy classification}}, AUTHOR = {Gan{\c c}arski, Pierre and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-03135400}, JOURNAL = {{Multimedia Tools and Applications}}, PUBLISHER = {{Springer Verlag}}, VOLUME = {35}, NUMBER = {1}, PAGES = {1-27}, YEAR = {2007}, HAL_ID = {hal-03135400}, HAL_VERSION = {v1}, } -
On machine learning in watershed segmentation
Conference IEEE International Workshop on Machine Learning in Signal Processing (MLSP), 2007, Greece. pp.187-192
Automatic image interpretation could be achieved by first performing a segmentation of the image, i.e. aggregating similar pixels to form regions, then use a supervised region- based classification. In such a process, the quality of the segmentation step is of great importance. Nevertheless, whereas the classification step takes advantage from some prior knowledge such as learning sample pixels, the segmentation step rarely does. In this paper, we propose to involve machine learning to improve the segmentation process using the watershed transform. More precisely, we apply a fuzzy supervised classification and a genetic algorithm in order to respectively generate the elevation map used in the watershed transform and tune segmentation parameters. The results from our evolutionary supervised watershed algorithm confirm the relevance of machine learning to introduce knowledge in the watershed segmentation process.
@inproceedings{derivaux:hal-00516076, TITLE = {{On machine learning in watershed segmentation}}, AUTHOR = {Derivaux, S{\'e}bastien and Lef{\`e}vre, S{\'e}bastien and Wemmert, C{\'e}dric and Korczak, Jerzy}, URL = {https://hal.science/hal-00516076}, BOOKTITLE = {{Proceedings of IEEE International Workshop on Machine Learning in Signal Processing (MLSP)}}, ADDRESS = {Greece}, PAGES = {187-192}, YEAR = {2007}, DOI = {10.1109/MLSP.2007.4414304}, PDF = {https://hal.science/hal-00516076v1/file/mlsp2007.pdf}, HAL_ID = {hal-00516076}, HAL_VERSION = {v1}, } -
Conference International Geoscience and Remote Sensing Symposium (IGARSS 2007), 2007, Barcelona, Spain. pp.3983 - 3986
Automatic image interpretation could be achieved by first performing a segmentation of the image, i.e. aggregating similar pixels to form regions, then use a region-based classification. This paper presents two region-based classifications, namely a supervised classification and an ontology-based classification and discuss their pros and cons. As they are complementary, we propose to combine these two approaches. Results shown that the presented method is relevant.
@inproceedings{derivaux:hal-01463171, TITLE = {{On the Complementarity of an Ontology and a Nearest Neighbour Classifier for Remotely Sensed Image Interpretation}}, AUTHOR = {Derivaux, S{\'e}bastien and Durand, Nicolas and Wemmert, C{\'e}dric}, URL = {https://hal.science/hal-01463171}, BOOKTITLE = {{International Geoscience and Remote Sensing Symposium (IGARSS 2007)}}, ADDRESS = {Barcelona, Spain}, PAGES = {3983 - 3986}, YEAR = {2007}, DOI = {10.1109/IGARSS.2007.4423093}, PDF = {https://hal.science/hal-01463171v1/file/ddw_igarss2007.pdf}, HAL_ID = {hal-01463171}, HAL_VERSION = {v1}, } -
Paramétrisation de méthodes de segmentation par utilisation de connaissances et approche génétique
Conference Atelier Extraction de COnnaissance à partir d'Images (ECOI), Journées Francophones Extraction et Gestion des Connaissances (EGC), 2007, Belgique in French
@inproceedings{derivaux:hal-00516103, TITLE = {{Param{\'e}trisation de m{\'e}thodes de segmentation par utilisation de connaissances et approche g{\'e}n{\'e}tique}}, AUTHOR = {Derivaux, S{\'e}bastien and Lef{\`e}vre, S{\'e}bastien and Wemmert, C{\'e}dric and Korczak, Jerzy}, URL = {https://hal.science/hal-00516103}, BOOKTITLE = {{Actes de l'Atelier Extraction de COnnaissance {\`a} partir d'Images (ECOI), Journ{\'e}es Francophones Extraction et Gestion des Connaissances (EGC)}}, ADDRESS = {Belgium}, PAGES = {-}, YEAR = {2007}, HAL_ID = {hal-00516103}, HAL_VERSION = {v1}, } -
Segmentation par ligne de partage des eaux basée sur des connaissances texturales
Conference Colloque GRETSI sur le Traitement du Signal et des Images, 2007, Troyes, France in French
La segmentation d'une image a pour but de créer des régions homogènes selon un critère défini en fonction de l'application considérée. Le critère peutêtre, par exemple, les valeurs des pixels de l'image ou des indices de texture. Une fois la segmentation obtenue, une classification des régions peutêtre effectuée afin d'associer une sémantique aux objets présents dans l'image. L'hypothèse d'homogénéité entre des pixels d'un même objet doit permettre la construction de régions correspondant aux objets de l'image. Dans le cas d'images complexes comme les images satellitesà très haute résolution spatiale, l'homogénéité classique des pixels d'un même objet n'est pas valide, réduisant la qualité des segmentations obtenues. Dans cet article, nous proposons d'intégrer au processus de segmentation par ligne de partage des eaux des connaissances sous forme d'exemplesétiquetés par l'utilisateur. Nous introduisons ainsi un critère d'homogénéité de plus haut niveau basé sur des connaissances texturales.
@inproceedings{derivaux:hal-00516100, TITLE = {{Segmentation par ligne de partage des eaux bas{\'e}e sur des connaissances texturales}}, AUTHOR = {Derivaux, S{\'e}bastien and Lef{\`e}vre, S{\'e}bastien and Wemmert, C{\'e}dric and Korczak, Jerzy}, URL = {https://hal.science/hal-00516100}, BOOKTITLE = {{Colloque GRETSI sur le Traitement du Signal et des Images}}, ADDRESS = {Troyes, France}, PAGES = {-}, YEAR = {2007}, PDF = {https://hal.science/hal-00516100v1/file/gretsi2007.pdf}, HAL_ID = {hal-00516100}, HAL_VERSION = {v1}, }
2006
-
Deriving Classification Rules from Multiple Remotely Sensed Data with Data Mining.
Conference 1st Workshop of the EARSeL Special Interest Group on Urban Remote Sensing, Mar 2006, Berlin, Germany. 9 p
@inproceedings{sheeren:halshs-00201917, TITLE = {{Deriving Classification Rules from Multiple Remotely Sensed Data with Data<br />Mining.}}, AUTHOR = {Sheeren, David and Puissant, Anne and Weber, Chrisitane and Gancarski, Pierre and Wemmert, C{\'e}dric}, URL = {https://shs.hal.science/halshs-00201917}, BOOKTITLE = {{Proceedings of the 1st Workshop of the EARSeL Special Interest Group on Urban Remote Sensing}}, ADDRESS = {Berlin, Germany}, PAGES = {9 p.}, YEAR = {2006}, MONTH = Mar, KEYWORDS = {Classification ; remotely Sensed Data ; Data mining}, HAL_ID = {halshs-00201917}, HAL_VERSION = {v1}, } -
Apport d'une classification non supervisée floue à la segmentation par ligne de partage des eaux
Conference Colloque International de Géomatique et d'Analyse Spatiale (SAGEO), 2006, France in French
@inproceedings{derivaux:hal-00516107, TITLE = {{Apport d'une classification non supervis{\'e}e floue {\`a} la segmentation par ligne de partage des eaux}}, AUTHOR = {Derivaux, S{\'e}bastien and Lef{\`e}vre, S{\'e}bastien and Wemmert, C{\'e}dric and Korczak, Jerzy}, URL = {https://hal.science/hal-00516107}, BOOKTITLE = {{Actes du Colloque International de G{\'e}omatique et d'Analyse Spatiale (SAGEO)}}, ADDRESS = {France}, PAGES = {-}, YEAR = {2006}, HAL_ID = {hal-00516107}, HAL_VERSION = {v1}, } -
Watershed Segmentation of Remotely Sensed Images Based on a Supervised Fuzzy Pixel Classification
Conference IEEE International Geosciences And Remote Sensing Symposium (IGARSS), 2006, United States. pp.3712-3715
Remotely sensed images are more and more precise (spatial resolution under 1 meter). For these images, objects of interest contains several pixels. Generally a segmentation method is used to cluster pixels that belong to the same objects before classification. The quality of such a segmentation method is crucial to achieve good clasification results. In this paper, a new segmentation method is proposed which aims to improve the classical watershed segmentation method based on multispectral gradient. The proposed method uses some labeled samples with classes of interest to induce a new dissimilarity between pixels which defines a new representation space to be used.
@inproceedings{derivaux:hal-00516090, TITLE = {{Watershed Segmentation of Remotely Sensed Images Based on a Supervised Fuzzy Pixel Classification}}, AUTHOR = {Derivaux, S{\'e}bastien and Lef{\`e}vre, S{\'e}bastien and Wemmert, C{\'e}dric and Korczak, Jerzy}, URL = {https://hal.science/hal-00516090}, BOOKTITLE = {{Proceedings of the IEEE International Geosciences And Remote Sensing Symposium (IGARSS)}}, ADDRESS = {United States}, PAGES = {3712-3715}, YEAR = {2006}, DOI = {10.1109/IGARSS.2006.951}, PDF = {https://hal.science/hal-00516090v1/file/igarss2006.pdf}, HAL_ID = {hal-00516090}, HAL_VERSION = {v1}, }
2002
-
A multi-view voting method to combine unsupervised classifications
Conference 2nd IASTED International Conference on Artificial Intelligence and Applications, 2002, Malaga, Spain. pp.447-452
@inproceedings{wemmert:hal-03135401, TITLE = {{A multi-view voting method to combine unsupervised classifications}}, AUTHOR = {Wemmert, C{\'e}dric and Gan{\c c}arski, Pierre}, URL = {https://hal.science/hal-03135401}, NOTE = {Artificial Intelligence and Applications}, BOOKTITLE = {{2nd IASTED International Conference on Artificial Intelligence and Applications}}, ADDRESS = {Malaga, Spain}, PAGES = {447-452}, YEAR = {2002}, HAL_ID = {hal-03135401}, HAL_VERSION = {v1}, }