Research 

On going research Project : SHAMI (SHApe Mutual Information)

Evaluation and fusion of segmentation methods in medical imaging

Summary of the project : In the field of delineation of 2D or 3D regions of interest (ROI) in medical imaging, and especially due to the development of multimodal and multiparametric image acquisition devices, the combination of segmentations of anatomical structures from different sources is of interest. It is also essential to accurately evaluate the variability between experts delineation or algorithms with different parameters. Finally, it seems important to be able to merge heterogeneous information from different modalities and acquisition methods in order to synthesize them. In this project, our aim is to propose the estimation of a reference shape defined as the optimum of a continuous criterion. We want to compare different shape metrics for segmentation evaluation and fusion and to compare these metrics to the well-known STAPLE approach. This work will also help us to evaluate the variability of deep learning methods and to combine segmentation results for more reproducibility.

keywords : Evaluation or fusion of segmentation methods in medical imaging, 2D and 3D Shape Analysis, Mean Shape, Mutual Shape, Mutual Information, shape gradients, level sets, deformable models

Some Publications 

ArXiv 2021 "A Mutual Reference Shape for Segmentation Fusion and Evaluation", S. Jehan-Besson, R. Clouard, C. Tilmant, A. de Cesare, A. Lalande, J. Lebenberg, P. Clarysse, L. Sarry, F. Frouin, M. Garreau,  arXiv:2102.08939, 2021

CS2022 "Optimization of a mutual shape based on the Fréchet-Nikodym metric for 3D shapes fusion". S.Jehan-Besson, P. Clarysse, R. Clouard, F. Frouin.  International Conference on Curves and Surfaces, Jun 2022, Arcachon, France. ⟨hal-03697975⟩

GSI2021 "Optimization of a shape metric based on information theory applied to segmentation fusion and evaluation in multimodal MRI for DIPG tumor analysis", S. Jehan-Besson, R. Clouard, N. Boddaert, J. Grill, F. Frouin, book chapter in LNCS, 5th International Conference on Geometric Science of Information GSI 2021, July Paris

Some presentations

FIL (Fédération Informatique Lyon) january 2023, "Optimization of a shape metric based on information theory for segmentation evaluation and fusion in 2D or 3D medical image analysis", seminar for new researchers in Lyon, 19 January 2023, links towards the pdf of the presentation

LABEX PRIMES december 2022, "Shape metrics for the evaluation of variability in deep learning segmentation methods for medical imaging", 8 december 2022. links towards the pdf of the presentation.

GSI  07-21 "Optimization of a shape metric based on information theory applied to segmentation fusion and evaluation in multimodal MRI for DIPG tumor analysis", S. Jehan-Besson, R. Clouard, N. Boddaert, J. Grill, F. Frouin, book chapter in LNCS, 5th International Conference on Geometric Science of Information GSI 2021, July Paris, link towards the conference paper in Springer, Link towards the pdf of the presentation GSI

Master 2 internship

Master 2, internship of A. Garcia, "Shape metrics for the evaluation and fusion of segmentation methods in multimodal 3D MRI", March 2023 - August 2023, CREATIS Laboratory, Lyon, under the supervision of S. Jehan-Besson and P. Clarysse.

Open source software under construction

ITK_MUTUAL_SHAPE 

From September 2023 to April 2024, we recruited A. Garcia as an engineer funded by the LABEX Primes (September 2023 to January 2024) and then funded by the CREATIS laboratory (January to April 2024) to provide a 2D, 3D and multi-labels mutual shape estimation module compatible with the ITK library, which is widely used in medical imaging. This module is a continuation of the work initiated by A. Garcia during his master's thesis and will be made available to the public in the near future. The multi-labels approach will make it possible to estimate mutual shapes composed of multiples labels in 2D or 3D. The theoretical work is available in the master thesis report of A. Garcia.

Summary of past research activities

Main studies :

• Design of 2D or 3D deformable models using shape gradients based on shape optimization theory using parametric and non parametric statistical criteria

• Segmentation fusion and evaluation without gold standard using a criterion based on information theory

• Color image filtering based on total variation and dual approaches

• Statistical inequalities for region matching.

• Applications : Medical Image Analysis (echocardiography, cardiac MRI, multi-modal brain MRI, perfusion cardiac MRI), Video image analysis, Color image filtering 

Some presentations that summarize these past research works  

GREYC 2017 "Approches variationnelles pour la segmentation, le filtrage et l'évaluation",  S. Jehan-Besson, Overview of research activities, January 2017, Link towards the pdf of the presentation

LITO Institut Curie 04-07 "Optimisation de critères statistiques pour la segmentation, l'évaluation ou la fusion de données, présentation de la méthodologie et étude des GITCs en IRM multimodale", S. Jehan-Besson, presentation within the LITO, 7 april 2021, Link towards the pdf of the presentation

GDR 06-13 "Approches variationnelles pour la segmentation : le point sur les critères à optimiser et les méthodes de résolution associées'', S. Jehan-Besson, invited speaker, Journées conjointes du GDR ISIS et du GDR Stic-Santé, Segmentation d'images médicales : approches variationnelles et statistiques, juin 2013, Link towards the pdf of presentation