Contact: matthieu.gilson [at] univ-amu.fr
PhD in neuroimaging biomarkers for psychiatry
Unlike neurological diseases, the characterization of psychiatric disorders remains at an early stage because several of them share the same symptoms. The identification of symptoms identified by clinicians during interviews are so far the standard to establish a diagnostic, but this empirical procedure is demanding as it must be repeated for follow-ups, and involve variability as it may vary depending on the clinicians’ own subjectivity. This calls for a better prediction for the diagnosis of patients, also for the disease evolution during a medication treatment. A promising direction concerns neuroimaging data that give access to both structure and activity of the brain.
The fusion of heterogeneous data (here functional and structural MRI, as well as clinical data) in automated analysis pipelines remains a crucial challenge in high-dimensional data analysis. The rapid advancements observed in various fields, such as image processing, natural language processing, and speech recognition, are largely attributable to the availability of large-scale datasets. However, in the medical domain, these approaches often encounter significant hurdles, primarily due to the relatively small size of patient cohorts and the high costs associated with data acquisition.
To address these issues, this project aims to benchmark existing methods from artificial intelligence (AI) and develop novel methods for multimodal classification that are suited for application in a clinical context. More precisely, it aims to improve the design of markers for bipolar disorder based on structural and functional MRI in combination with clinical data (cognitive scores, etc.), including both patients and healthy controls.
The goal of the PhD project is to develop computational tools for diagnosis and prognosis, like predicting the evolution of a patient subject to medication or patient stratification to identify subtypes of psychiatric diseases.
The objectives of the present project are thus focused on data analysis:
The PhD candidate will be trained to acquire the different techniques required for the project, ranging from data preprocessing to machine-learning tools, as well as data acquisition to some extent.
- development of multimodal fusion pipeline that combines the heterogeneous data for classification,
- testing self-supervised representational learning approach for the high-dimensional multivariate data (MRI).
The multidisciplinary supervision panel has complementary expertise covering the fields of computational neuroscience, machine learning and medicine:
- Emmanuel Daucé (emmanuel.dauce.free.fr)
- Antoine Lefrère (https://www.researchgate.net/profile/Antoine-Lefrere)
- Matthieu Gilson (https://matthieugilson.eu)
computer science, biomedical, neuroscience, signal processing
Skills to acquire:
neuroscience: neuroimaging (functional and structural MRI), brain networks
machine learning: multimomdal classification, representational learning
computer science: programming in Python, high-performance computing
Dabane G, Perrinet LU, Daucé E (2022) What You See Is What You Transform: Foveated Spatial Transformers as a bio-inspired attention mechanism, 2022 International Joint Conference on Neural Networks (IJCNN), 1-8
Adhikari MH, ..., Deco G, Instabato A, Gilson M**, Corbetta M** (2021) Effective connectivity extracts clinically relevant prognostic information from resting state activity in stroke. Brain Commun 3: fcab233
Kobeleva X, Varoquaux G, ..., Grefkes C, Gilson M. (accepted) Advancing brain network models to reconcile functional neuroimaging and clinical research. Neuroimage Clin 36: 103262
Lefrere A, Auzias G, Favre P, Kaltenmark I, Houenou J, Piguet C, et al. (2023) Global and local cortical folding alterations are associated with neurodevelopmental subtype in bipolar disorders: a sulcal pits analysis. J Affect Disord 325: 224-230
Dadi K, ..., Thirion B, Varoquaux G, Alzheimer's Disease Neuroimaging Initiative (2019) Benchmarking functional connectome-based predictive models for resting-state fMRI. Neuroimage 192: 115-134