DeepLesionBrain: Towards a broader deep-learning generalization for multiple sclerosis lesion segmentation
| dc.rights.license | open | en_US |
| hal.structure.identifier | Laboratoire Bordelais de Recherche en Informatique [LaBRI] | |
| dc.contributor.author | KAMRAOUI, Reda Abdellah | |
| hal.structure.identifier | Laboratoire Bordelais de Recherche en Informatique [LaBRI] | |
| dc.contributor.author | TA, Vinh-Thong | |
| hal.structure.identifier | Neurocentre Magendie : Physiopathologie de la Plasticité Neuronale [U1215 Inserm - UB] | |
| dc.contributor.author | TOURDIAS, Thomas | |
| hal.structure.identifier | Laboratoire Bordelais de Recherche en Informatique [LaBRI] | |
| dc.contributor.author | MANSENCAL, Boris
IDREF: 228223601 | |
| hal.structure.identifier | Universitat Politècnica de València = Universitad Politecnica de Valencia = Polytechnic University of Valencia [UPV] | |
| dc.contributor.author | MANJON, Jose V. | |
| hal.structure.identifier | Laboratoire Bordelais de Recherche en Informatique [LaBRI] | |
| dc.contributor.author | COUPE, Pierrick | |
| dc.date.accessioned | 2023-01-06T09:06:14Z | |
| dc.date.available | 2023-01-06T09:06:14Z | |
| dc.date.created | 2022-06-13 | |
| dc.identifier.uri | https://oskar-bordeaux.fr/handle/20.500.12278/171607 | |
| dc.description.abstractEn | Recently, segmentation methods based on Convolutional Neural Networks (CNNs) showed promising performance in automatic Multiple Sclerosis (MS) lesions segmentation. These techniques have even outperformed human experts in controlled evaluation conditions such as Longitudinal MS Lesion Segmentation Challenge (ISBI Challenge). However, state-of-the-art approaches trained to perform well on highly-controlled datasets fail to generalize on clinical data from unseen datasets. Instead of proposing another improvement of the segmentation accuracy, we propose a novel method robust to domain shift and performing well on unseen datasets, called DeepLesionBrain (DLB). This generalization property results from three main contributions. First, DLB is based on a large group of compact 3D CNNs. This spatially distributed strategy aims to produce a robust prediction despite the risk of generalization failure of some individual networks. Second, we propose a hierarchical specialization learning (HSL) by pre-training a generic network over the whole brain, before using its weights as initialization to locally specialized networks. By this end, DLB learns both generic features extracted at global image level and specific features extracted at local image level. Finally, DLB includes a new image quality data augmentation to reduce dependency to training data specificity (e.g., acquisition protocol). DLB generalization was validated in cross-dataset experiments on MSSEG’16, ISBI challenge, and in-house datasets. During experiments, DLB showed higher segmentation accuracy, better segmentation consistency and greater generalization performance compared to state-of-the-art methods. Therefore, DLB offers a robust framework well-suited for clinical practice. © 2021 | |
| dc.description.sponsorship | Apprentissage profond pour la volumétrie cérébrale : vers le BigData en neuroscience - ANR-18-CE45-0013 | en_US |
| dc.description.sponsorship | Translational Research and Advanced Imaging Laboratory - ANR-10-LABX-0057 | en_US |
| dc.description.sponsorship | Initiative d'excellence de l'Université de Bordeaux - ANR-10-IDEX-0003 | en_US |
| dc.language.iso | EN | en_US |
| dc.subject.en | Deep Learning | |
| dc.subject.en | Convolutional Neural Network | |
| dc.subject.en | Domain Generalization | |
| dc.subject.en | Learning Generalization | |
| dc.subject.en | Lesion Segmentations | |
| dc.subject.en | Multiple Sclerose Segmentation | |
| dc.subject.en | Multiple Sclerosis | |
| dc.subject.en | Multiple Sclerosis Lesions | |
| dc.subject.en | Segmentation Accuracy | |
| dc.subject.en | Convolutional Neural Networks | |
| dc.subject.en | Convolutional Neural Network | |
| dc.subject.en | Feature Extraction | |
| dc.subject.en | Image Quality | |
| dc.subject.en | Image Segmentation | |
| dc.subject.en | Prediction | |
| dc.subject.en | Three-Dimensional Imaging | |
| dc.subject.en | Brain | |
| dc.subject.en | Diagnostic Imaging | |
| dc.subject.en | Image Processing | |
| dc.subject.en | Pathology | |
| dc.subject.en | Procedures | |
| dc.subject.en | Humans | |
| dc.subject.en | Computer-Assisted | |
| dc.subject.en | Neural Networks | |
| dc.title.en | DeepLesionBrain: Towards a broader deep-learning generalization for multiple sclerosis lesion segmentation | |
| dc.title.alternative | Med Image Anal | en_US |
| dc.type | Document de travail - Pré-publication | en_US |
| dc.identifier.doi | 10.1016/j.media.2021.102312 | en_US |
| dc.subject.hal | Sciences du Vivant [q-bio]/Neurosciences [q-bio.NC] | en_US |
| bordeaux.page | 102312 | en_US |
| bordeaux.hal.laboratories | Neurocentre Magendie - U1215 | en_US |
| bordeaux.institution | Université de Bordeaux | en_US |
| bordeaux.institution | INSERM | en_US |
| bordeaux.institution | Bordeaux INP | |
| bordeaux.institution | CNRS | |
| bordeaux.team | Relations glie-neurone | en_US |
| bordeaux.identifier.funderID | Ministerio de Economía y Competitividad | en_US |
| hal.export | false | |
| dc.rights.cc | Pas de Licence CC | en_US |
| bordeaux.COinS | ctx_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.spage=102312&rft.epage=102312&rft.au=KAMRAOUI,%20Reda%20Abdellah&TA,%20Vinh-Thong&TOURDIAS,%20Thomas&MANSENCAL,%20Boris&MANJON,%20Jose%20V.&rft.genre=preprint |
