Deep Learning methods for fetal cortical surfaces generation from MRI scans

When:
15/01/2024 all-day
2024-01-15T01:00:00+01:00
2024-01-15T01:00:00+01:00

Offre en lien avec l’Action/le Réseau : – — –/– — –

Laboratoire/Entreprise : Institut de Neurosciences de la Timone
Durée : 5 ou 6 mois
Contact : olivier.coulon@univ-amu.fr
Date limite de publication : 2024-01-15

Contexte :
The human cerebral cortex undergoes dynamic and regionally heterogeneous development during gestation [1]. Cortical surface-based analysis is based on the reconstruction of topologically-correct and geometrically accurate surface representations of the folded, thin cerebral cortex. The fetal cerebral cortex is typically represented as a triangulated mesh with a spherical topology for each brain hemisphere, generated from a tissue segmentation of the structural MRI scan.

Sujet :
In this context, and although various algorithmic solutions have been proposed in the past to propose such meshes, deep learning approaches have now become state-of-the-art for cortical surface generation [2]. The goal of this internship is to evaluate some of these methods. In particular we will evaluate their performance on fetal MR data, that are difficult to process with standard methods.
The candidate will be in charge of:
1. Testing pre-selected deep-learning approaches from the literature (CortexODE [3] CorticalFlow++[4], DeepCSR[5], PialNN[6], Topofit[7]).
2. Comparing these methods on a large dataset of fetal MRI available in the team and containing normal and pathological scans
3. Perform an application study on the obtained surfaces representations to characterize normal and pathological fetal development (optional, depending on progress)

[1] I.Kostović,G.Sedmak,andM.Judaš,“Neuralhistologyandneurogenesisofthehumanfetal and infant brain,” NeuroImage. 2019, doi: 10.1016/j.neuroimage.2018.12.043.
[2] F.Zhao,Z.Wu,andG.Li,“Deeplearningincorticalsurface-basedneuroimageanalysis:a systematic review,” Intell. Med., 2023, doi: 10.1016/j.imed.2022.06.002.
[3] Q.Ma,L.Li,E.C.Robinson,B.Kainz,D.Rueckert,andA.Alansary,“CortexODE:Learning Cortical Surface Reconstruction by Neural ODEs.” arXiv, 2022 http://arxiv.org/abs/2202.08329
[4] R.SantaCruzetal.,“CorticalFlow++:BoostingCorticalSurfaceReconstructionAccuracy, Regularity, and Interoperability,” in MICCAI 2022, doi: 10.1007/978-3-031-16443-9_48.
[5] R.S.Cruz,L.Lebrat,P.Bourgeat,C.Fookes,J.Fripp,andO.Salvado,“DeepCSR:A3D Deep Learning Approach for Cortical Surface Reconstruction,” IEEE/CVF, 2021, https://openaccess.thecvf.com/content/WACV2021/html/Santa_Cruz_DeepCSR_A_3D_Deep_ Learning_Approach_for_Cortical_Surface_Reconstruction_WACV_2021_paper.html
[6] Q.Ma,E.C.Robinson,B.Kainz,D.Rueckert,andA.Alansary,“PialNN:AFastDeepLearning Framework for Cortical Pial Surface Reconstruction,” in Machine Learning in Clinical Neuroimaging, 2021, doi: 10.1007/978-3-030-87586-2_8.
[7] A.Hoopes,J.E.Iglesias,B.Fischl,D.Greve,andA.V.Dalca,“TopoFit:RapidReconstruction of Topologically-Correct Cortical Surfaces,” MIDL, 2021, https://openreview.net/forum?id=-JiHeZNDY3a

Profil du candidat :
Nous recherchons un étudiant en M2 ou Dernière année d’école d’ingénieur, motivé par l’imagerie médicale ou les neurosciences, et possédant les compétences suivantes:

Formation et compétences requises :
– une bonne connaissance des principes de l’apprentissage profond.
– une expérience de python pour l’apprentissage profond (e.g. Pytorch)
– une connaissance de Git et des environnements de type Linux.
– une expériences précédente dans le domaine de l’apprentissage profond est un plus.

Adresse d’emploi :
The intern will integrate the MeCA research team of the Institut des Neurosciences de la Timone, in Marseille, France. The Meca team (www;meca-brain.org)combines expertise in processing of large fetal MRI databases and surface based morphometry methods. Tools and data required for the internship will be provided by the team.

Document attaché : 202310201221_2024_M2_internship_surface_extraction.pdf