The Image Processing and Pattern Recognition group has been involved in medical imaging since 2017. The team primarily focuses on detection and segmentation tasks. Building on its expertise in mathematical morphology, the team designs robust and efficient algorithms that integrate morphological tools into deep learning models and that can handle data disparities such as those introduced by varying acquisition centers and modalities. These algorithms are designed to detect small, low-contrast structures and run efficiently, making them suitable for routine clinical use. In addition, the team regularly participates to medical imaging challenges organized by MICCAI and ISBI conferences, and has scored several notable results.
Our contributions in medical image analysis
The research conducted in the group has been particularly directed toward the following anatomical structures and organs:
- Brain with the segmentation of brain structures (in neonates, isointense phase, and adults) as well as the detection and segmentation of lesions (brain tumors, vascular lesions) [3][1].
- Heart with the segmentation of various structures, such as the ventricles, left atrium, and myocardium, for cardiac fibrosis prediction [4][5][6].
- Blood vessels with the segmentation of vascular networks and the definition of robust evaluation metrics accounting for uncertainties in ground truth accuracy.
For these tasks, the team focuses on developing fast and robust pipelines. A key and original contribution has been the adaptation of the VGG16 model for segmenting 3D medical images by concatenating three slices into a single 2D color image. This technique takes advantage of 3D information and can be applied to consecutive slices, different modalities, or a combinations of both [2]. The integration of morphological pre-processing has also led to notable improvements in results, particularly for detecting and segmenting fine lesions.
The team also contributes to the development of the open-source software Holovibes, led by Michael Atlan (associate researcher in the Image Processing and Pattern Recognition group). Holovibes is a free software dedicated to the calculation of holograms in real-time for Doppler imaging and optical coherence tomography in ophthalmology.
Our participation to medical imagery challenges
Since 2017, the Image Processing and Pattern Recognition group has participated to several medical imaging challengesorganized by MICCAI and ISBI conferences. These challenges provide an opportunity to develop original methods capitalizing on the team strengths (particularly in the field of mathematical morphology) and serve as catalysts for the team dynamics.
The team participation has regularly been awarded, both for the achieved high quantitative results and the originality of the developed methods.
Challenges where the team results have been awarded (top 5)
- AtriaSeg 2018 : 3rd place for the automatic segmentation of the left atrium cavity.
- BraTS 2018 : 2nd place for the survival prediction task.
- BraTS 2019 : 5th place for the survival prediction task.
- LVQuan 2019 : 3rd place for the left ventricle full quantification
- Valdo 2021 : 1st place for the segmentation of enlarged perivascular spaces.
Related Projects
DeepToS
The DeepToS project explores a novel approach to 3D vascular segmentation by classifying nodes in a Tree of Shapes structure using deep learning and graph-based methods, with the goal of producing more connected and biologically consistent results.
ISEVAC
The ISEVAC project develops next-generation vascular segmentation methods by combining high-quality annotations, topology-aware tracking, and foundation-model-guided interaction, with direct integration into 3D Slicer.
Related Publications
[1]
Élodie Puybareau • Guillaume Tochon • Joseph Chazalon • Jonathan Fabrizio. "Segmentation of Gliomas and Prediction of Patient Overall Survival: A Simple and Fast Procedure". Proceedings of the Workshop on Brain Lesions (BrainLes), in conjunction with MICCAI. 2018. https://doi.org/10.1007/978-3-030-11726-9_18.
[2]
Yongchao Xu • Thierry Géraud • Isabelle Bloch. "From Neonatal to Adult Brain MR Image Segmentation in a Few Seconds Using 3D-Like Fully Convolutional Network and Transfer Learning". Proceedings of the 23rd IEEE International Conference on Image Processing (ICIP). 2017. https://doi.org/10.1109/ICIP.2017.8297117.
[3]
Yongchao Xu • Baptiste Morel • Sonia Dahdouh • Élodie Puybareau and Alessio Virzì • Hélène Urien • Thierry Géraud • Catherine Adamsbaum • Isabelle Bloch. "The Challenge of Cerebral Magnetic Resonance Imaging in Neonates: A New Method using Mathematical Morphology for the Segmentation of Structures Including Diffuse Excessive High Signal Intensities". Medical Image Analysis. 2018. https://doi.org/10.1016/j.media.2018.05.003.
[4]
Zhou Zhao • Nicolas Boutry • Élodie Puybareau • Thierry Géraud. "A Two-Stage Temporal-Like Fully Convolutional Network Framework for Left Ventricle Segmentation and Quantification on MR Images". Statistical Atlases and Computational Models of the Heart. Multi-Sequence CMR Segmentation, CRT-EPiggy and LV Full Quantification Challenges---10th International Workshop, STACOM 2019, Held in Conjunction with MICCAI 2019, Shenzhen, China, October 13, 2019, Revised Selected Papers. 2020. https://doi.org/10.1007/978-3-030-39074-7_42.
[5]
Zhou Zhao • Nicolas Boutry • Élodie Puybareau • Thierry Géraud. "FOANet: A Focus of Attention Network with Application to Myocardium Segmentation". Proceedings of the 25th International Conference on Pattern Recognition (ICPR). 2021. https://doi.org/10.1109/ICPR48806.2021.9412016.
[6]
Zhou Zhao • Nicolas Boutry • Élodie Puybareau • Thierry Géraud. "Do not Treat Boundaries and Regions Differently: An Example on Heart Left Atrial Segmentation". Proceedings of the 25th International Conference on Pattern Recognition (ICPR). 2021. https://doi.org/10.1109/ICPR48806.2021.9412755.