Fast matching in fingerprints dictionary using deep learning : application to estimation of brain microstructure

Details

Supervisors

Benoit Macq

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en mathématiques appliquées, à finalité spécialisée

Abstract

The impact and interest of deep learning has been growing recently and has empirically shown great precision and efficiency performance. Especially feed-forward networks are very fast to evaluate once trained and can theoretically learn any input-output mapping. This work investigates different methodologies to accelerate very significantly a slow but easily interpretable dictionary search, at the cost of a loss of interpretability. We concentrate particularly on the use of neural networks, which in this work perform the best in both precision and time efficiency, but also depict a more “black-box” character. This master thesis focuses on the problem of brain microstructure estimation using MRI signals and a dictionary of Dw-MRI fingerprints, with the goal to advance our knowledge of psychiatric and neurological disorders.