French
English
Scientific context
The development of multimodal machine learning models for medical data analysis is a recent and growing field of research [1]. Nevertheless, combining heterogeneous data, such as image and signal data remains challenging, and is typically achieved by combining single-modality models’ decisions. An innovative solution relies on graph-based models, which can be used to represent heterogeneous data in a common space [2]. Graph structure is specifically adapted to represent brain data, where nodes and edges can represent brain regions of interest and connectivity, respectively. In addition, graph structure can be used as input to machine learning models. Building multimodal graph models involve two main steps: graph construction, including feature extraction and node and edge definition, followed by training of the graph-based model.
In the context of this internship, we will focus on two specific modalities: structural magnetic resonance imaging (image data - MRI) and magnetoencephalography (multivariate time-series data - MEG), which captures neural brain activity. The data will be used to train a model to estimate the biological brain age. Biological brain age estimation is a highly clinically relevant task [3], and for which multimodality has been shown to improve prediction accuracy [4]. In healthy subjects, we hypothesize that the biological brain age is equal to the chronological age, which is used as the ground truth for model training. When further tested on general population data, the brain age estimated by the model can be compared to the chronological age of the subject, and a significant difference between these values can be indicative of several neurological disorders.
Objectives of the internship
The Master student will use open data from the Cam-CAN database (Cam-CAN Data Repository), which contains structural MRI and resting-state MEG recordings from ~650 healthy adults (aged 18-90 year old).
The objective of the internship is twofold:
- The first part will involve preprocessing of the MRI and MEG data, feature extraction and subject-level multimodal graphs construction. The multimodal graphs will embed cortical imaging [5] and neural activity features [6], with ROIs representing the different cortical structures. Thus, image segmentation pipelines (Freesurfer) and signal analysis methods (mne-Python) will be used to derive features from the raw data.
- The second part will involve exploring several graph neural network (GNN) architectures to train a regression model for biological brain age estimation directly from the graph, using the Pytorch Geometric library. The benefit of using a graph structure will be assessed by comparing GNN model results to a traditional machine learning model working directly in the feature space.
The intern will benefit from the expertise in image processing and deep learning from the Myriad team members. In addition, the intern will be invited to join the collaborative ANR SEIZURE project meetings, led by C. Lartizien, which focuses on developing multimodal models from imaging and electrophysiology data in the context of epilepsy surgical planning. Thus, the intern will benefit from the interdisciplinary expertise of the ANR SEIZURE project members.
Skills
The Master’s student should have a background in image and signal processing and/or deep learning. Proficiency in Python programming is required and a strong interest in clinical neuroscience would be appreciated.
Supervision
Pauline Mouchès and Carole Lartizien
Application
Interested applicants are required to send a cover letter, CV and any other relevant documents (reference letter, recent transcripts of marks, ...) to pauline.mouches@inserm.fr
Location: Laboratoire CREATIS, campus de la DOUA, Villeurbanne (Lyon)
Gratification: ~650 €/month
Duration: ~6 months, starting from january/february
Language: French or English
References
[1] Kline, A., et al., 2022. Multimodal machine learning in precision health: A scoping review. npj Digit. Med. 5, 1–14.
[2] Ding, K., et al., 2022. Graph Convolutional Networks for Multi-modality Medical Imaging: Methods, Architectures, and Clinical Applications.
[3] Baecker, L., et al., 2021. Machine learning for brain age prediction: Introduction to methods and clinical applications. eBioMedicine 72.
[4] Cole, J.H., 2020. Multimodality neuroimaging brain-age in UK biobank: relationship to biomedical, lifestyle, and cognitive factors. Neurobiology of Aging 92, 34–42.
[5] Chen, E., et al., 2024. Multiple sclerosis clinical forms classification with graph convolutional networks based on brain morphological connectivity. Front. Neurosci. 17.
[6] Mouches, P., et al., 2024. Time CNN and Graph Convolution Network for Epileptic Spike Detection in Meg Data, in: 2024 IEEE International Symposium on Biomedical Imaging (ISBI).