Thèse Cartographie des Lésions de la Substance Blanche et Analyse des Déconnexions Structurelles et Fonctionnelles chez des Patients en Phase Chronique d'Avc par Irm Multimodale H/F - Doctorat.Gouv.Fr

Établissement : Université Paris-Saclay GS Sciences de l'ingénierie et des systèmes École doctorale : Electrical, Optical, Bio-physics and Engineering Laboratoire de recherche : Modèles et inférence pour les données de Neuroimagerie Direction de la thèse : Philippe CIUCIU ORCID 000000015374962X Début de la thèse : 2026-12-01 Date limite de candidature : 2026-10-10T23:59:59 La récupération motrice après un accident vasculaire cérébral (AVC) ischémique repose sur une réorganisation des réseaux cérébraux encore mal comprise. Dans le cadre du programme BrainSync, l'étude MOTIF-STROKE combine l'IRM multimodale à 7 T et à 3 T (IRM quantitative, IRM de diffusion, IRMf de repos et IRMf de tâche) à des évaluations motrices cliniques et numériques. Elle porte sur des patients victimes d'AVC présentant une atteinte motrice du membre supérieur.
Cette thèse porte sur la dimension structurelle de l'atlas MOTIF-STROKE des déconnexions post-AVC et sur ses liens avec la fonction et le comportement. Elle s'articule autour de trois projets :
1. Introduire la notion de « passabilité » (ou perméabilité) lésionnelle pour montrer que la microstructure de la lésion, et non seulement sa topographie, façonne la déconnexion structurelle individuelle.
2. Confronter les capacités structurelle et fonctionnelle des patients, prédites à partir de la connectivité, pour identifier des sous-groupes au potentiel de récupération distinct.
3. Extraire des modes latents reliant la déconnexion à l'échelle du cerveau entier au remodelage des activations motrices induites par la tâche.

Les biomarqueurs d'imagerie interprétables qui en résulteront aideront à prédire la récupération et à personnaliser la rééducation. Ils serviront également à cibler l'implantation du dispositif WIMAGINE dans le futur essai BCI4STROKE. Ischemic stroke is a leading cause of long-term motor disability. Functional recovery relies on a complex reorganization of brain networks that remains incompletely understood [Favre 2014; Hommel 2016]. The BrainSync project (November 2024 - May 2029), funded by the CEA high-risk research program Audace!, aims to develop innovative interventional strategies for upper-limb motor rehabilitation after stroke.
Within BrainSync, the MOTIF-STROKE study combines ultra-high-field (7 T) and high-field (3 T) MRI, artificial intelligence and digital consultations to map motor intentions, characterize brain lesions and quantify motor disability in stroke patients with upper-limb impairment. Its overarching aim is to build a high-resolution morphological, functional and structural atlas of post-stroke disconnection and reorganization. Building on the disconnectome framework [Thiebaut de Schotten 2020; Forkel 2022], this atlas will couple lesion-derived structural disconnection maps with multimodal MRI data to decode the relationships between anatomical disconnections, functional connectivity alterations and behavioral outcomes.
Prior lesion-function studies in stroke have typically been limited to small cohorts (20-25 patients), low-resolution 3 T MRI and a lack of integration between structural (diffusion) and functional (resting-state fMRI) connectivity [Jaillard 2005; Hannanu 2017; Hannanu 2020]. Moreover, lesions are usually treated as binary masks: normative disconnection methods [Kuceyeski 2013; Thiebaut de Schotten 2020] infer which tracts are interrupted from the lesion location only, ignoring the substantial microstructural heterogeneity of chronic lesions (e.g., CSF-like versus gray-matter-like tissue [Krishnamurthy 2021]), which can determine whether fibers actually remain intact through damaged tissue, as shown in glioma [Hu 2025]. Likewise, structural and functional connectivity have each been shown to predict motor outcome [Yu 2023; Zhao 2023], but how these two sources of information interact - and what their disagreement reveals about recovery - is unknown. Finally, focal studies have related specific tracts (e.g., the corpus callosum) to the laterality of motor activations [Wang 2012], but how whole-brain disconnection patterns drive specific functional remapping remains unclear. The overall objective is to elucidate the relationships between white-matter lesions, structural and functional disconnections, and motor disability, at both the individual and cohort levels. In line with the four MOTIF-STROKE goals - (i) mapping, (ii) association, (iii) mechanism and (iv) prediction - the thesis pursues three specific objectives:
- Objective 1 - Beyond topography (mapping). Link lesion microstructure to individualized structural connectivity (SC): define and quantify lesion passability from each patient's diffusion MRI, characterize its microstructural correlates (axonal density, free water, myelin, iron), and test whether individualized SC captures post-stroke disconnection beyond normative, overlap-based predictions.
- Objective 2 - Structural preservation vs. functional reorganization (association, mechanism, prediction). Quantify each patient's structural capacity (preserved structural substrate) and functional capacity (effective functional reorganization) through motor-performance prediction, and determine whether their alignment or mismatch defines subgroups with distinct network features and recovery trajectories.
- Objective 3 - Disconnection-constrained functional remapping (mapping, association, mechanism). Identify data-driven, whole-brain modes coupling structural disconnection patterns with task-evoked motor remapping at 7 T, and assess their network-level interpretation and behavioral relevance.
1 Cohort and data
The thesis relies on the MOTIF-STROKE cohort acquired at NeuroSpin: ischemic stroke patients with motor impairment, systematically recruited across diverse vascular territories to cover the range of disconnection patterns relevant to motor recovery (target: 100 patients), and 25 healthy controls. Data include:
- 7 T MRI: MP2RAGE (T1-weighted images and quantitative R1 maps), quantitative R2* maps, and task fMRI (four-limb motor movement localizer, including hand and foot movements, and an orientation task).
- 3 T MRI: MP2RAGE, FLAIR, multi-shell diffusion-weighted imaging (DWI) and resting-state fMRI.
- Clinical and behavioral data (baseline and follow-up): Fugl-Meyer assessment and digital consultations extracting objective indices of walking (e.g., gait asymmetry) and object grasping (lateralized grasping performance) [Jaillard 2021].

2 Image processing
Lesions will be delineated on high-resolution FLAIR and T1-weighted contrasts and projected into MNI space [Brett 2002]. Diffusion data will be denoised [Veraart 2016], modeled with multi-shell multi-tissue constrained spherical deconvolution [Jeurissen 2014] for whole-brain tractography, and fitted with NODDI to estimate neurite density (NDI) and isotropic free-water fraction (ISOVF) [Zhang 2012]. Quantitative R1 and R2* maps provide myelin- and iron-sensitive measures. fMRI data will be denoised with local low-rank methods [Comby 2023] and analyzed with standard GLM and connectivity pipelines [Abraham 2014]. Superficial and deep white-matter bundle atlases developed at NeuroSpin [Chauvel 2024; Herlin 2023] will support tract-level analyses.

3 Project 1 - Lesion passability and individualized structural connectivity
- Passability by tract t in patient i is defined as P(i,t) = N_pass(i,t) / N_streamline(i,t), i.e., the fraction of streamlines entering a lesion that pass through it; an overall passability index is obtained as a weighted sum across tracts, P(i) = _t w(i,t) P(i,t).
- Normative (topography-based) disconnection is computed by mapping each lesion onto a normative structural connectome: predicted loss for an edge = number of streamlines overlapping the lesion / total number of streamlines [Kuceyeski 2013; Thiebaut de Schotten 2020].
- Actual loss is measured as the deviation (z-score) of the patient's individualized SC from the control group. Nested models (Actual loss ~ Predicted loss, versus Actual loss ~ Passability + Predicted loss + Passability × Predicted loss) will quantify the gain in explained variance (R²) brought by lesion microstructure.
- Passability will be validated against multimodal microstructural markers: axonal density (NDI) and free water (ISOVF) from NODDI, myelin (R1) and iron (R2*) from quantitative MRI.

4 Project 2 - Structural and functional capacity
- Motor performance will be predicted separately from SC and from resting-state FC using leave-one-patient-out cross-validated models; predicted scores define each patient's structural capacity (S) and functional capacity (F).
- Patients will be placed in a capacity alignment space (low/high S × low/high F). Mismatched quadrants are of particular interest: high S / low F (inefficient reorganization, high recovery potential) and low S / high F (constrained functional compensation).
- Subgroups will be compared on SC, FC and structure-function coupling features and on motor outcomes, both at baseline and longitudinally, to test whether preserved structural substrate imposes a stronger constraint on recovery than functional compensation alone.

5 Project 3 - Disconnection-constrained functional remapping
- Following ROI and edge selection, structural disconnection (X) and motor task-evoked activation from the 7 T localizer (Y) will be expressed as z-scores relative to controls.
- Partial least squares (PLS) [Krishnan 2011] will extract orthogonal latent variables (LV1, LV2, ...), each defined by a disconnection pattern (X loadings), a remapping pattern (Y loadings) and patient-specific scores.
- Latent modes will be interpreted with network metrics (e.g., communicability to the somatomotor, frontoparietal, dorsal attention and default mode networks) and related to motor performance (e.g., an interhemispheric compensation mode).

Le candidat est déjà identifié, il s'agit de Chengyi Yuan, de nationalité chinoise (dossier FSD validé côté Inria), qui a fait tout le parcours de sélection du DIM C-BRAIN en 2025/2026, et a été interviewé mais n'a pas obtenu le financement.