Soutenance de thèse de Rohit Yadav, MeCA team

Thursday December 14, 2023, at 9:00 AM, Salle de Thèse 2, Faculté de Médecine

Rohit Yadav, MeCA team (Méthodes et Anatomie Computationnelle)
Development of normative and pathological models of cortical folding by machine learning on graphs

Jury:
Jean François MANGIN –  Directeur de recherche – Rapporte

Luc BRUN – Professeur, ENSICAEN – Rapporteur

Carole LARTIZIEN – Directrice de recherche, CNRS – Examinatrice 

Sophie ACHARD – Directrice de recherche, CNRS – Examinatrice

Benoit GAÜZÈRE – Maître de conférence, INSA ROUEN – Examinateur

Olivier COULON –  Directeur de  recherche, CNRS –  Directeur de thèse

François-Xavier DUPÉ – Maître de conférence, Aix-Marseille Université – Co-directeur de thèse

Guillaume AUZIAS – Chargé de recherche, CNRS  – Co-encadrant de thèse

Abstract: The human cerebral cortex is remarkable for its complex, tortuous geometry of fissures called sulci. The global network of cortical sulci across the brain surface provides important insights into the processes of brain development and functional organisation. Sulcal patterns have served as critical landmarks that experts rely on to assess potential abnormalities in cerebral structure. These patterns not only aid early intervention for cerebral malformations, but also contribute to our broader understanding of how the human brain develops. However, the patterns of cortical folding, i.e. the arrangement, characteristics and shape of sulci, are unique to each individual, making the identification of biomarkers from abnormal patterns both methodologically and conceptually challenging. Traditional registration-based techniques often treat these variations as noise, resulting in only an implicit matching of folds. An alternative approach that has emerged in recent years is to explicitly extract and identify cortical folds and model them as graphs of interconnected sulcal basins. This provides a systematic perspective to study the organisation of these folds, allowing us to examine folding patterns in terms of graph structure. Combining these graph-based modelling approaches with machine learning methods should provide interesting insights into understanding inter-individual variation and similarities in network patterns, including local regions. The main aim of this thesis is to propose new tools to quantify the variations in these graphs representing individual folding patterns.