French
English
Please see attached file for more information
Key words: Machine learning, Deep Learning; Multidimensional data, Segmentation, Neuroimaging, Self-supervised learning, Anomaly detection, Unsupervised representation learning
Scientific context
The vast majority of deep learning architectures for medical image analysis are based on supervised models requiring the collection of large datasets of annotated examples. Building such annotated datasets, which requires skilled medical experts, is time consuming and hardly achievable, especially for some specific tasks, including the detection of small and subtle lesions that are sometimes impossible to visually detect and thus manually outline. This critical aspect significantly impairs performances of supervised models and hampers their deployment in clinical neuroimaging applications, especially for brain pathologies that require the detection of small size lesions (e.g. multiple sclerosis, microbleeds) or subtle structural or morphological changes (e.g. Parkinson disease).
Objective and research program
To solve this challenging issue, the objective of this thesis is to develop and evaluate deep self-supervised detection and segmentation approaches whose training does not require any fine semantic annotations of the anomalies localization. We will explore different categories of self-supervised methods, including : novel unsupervised auto-encoder based anomaly detection models leveraging on the recent developments in visual transformers blocks (ViT) or vector quantized variational autoencoders (VQ-VAE), scalability of Gaussian mixture models as well as weakly supervised models based on scarce annotations.
In a first step, we will focus on Parkinson disease and micro hemorrhage imaging data and fuse different MR modalities.
Environment: The PhD position is granted by the “Défi IA” program sponsored by la Région Auvergne Rhône-Alpes. We offer a stimulating research environment gathering experts in Image processing, Neurosciences & Neuroimaging, Advanced Statistical and Machine Learning methods from CREATIS, Grenoble Institute of Neurosciences (GIN) and INRIA.
How to apply: Send an email directly to the three supervisors with your CV and persons to contact. Interviews of the selected applicants will be done on an ongoing basis. Applications will be accepted up to the 30st of June.