The Artificial Intelligence Slot. Overview of Neural Networks and application to diffusion weighted MRI.

Nowadays deep learning based methods are being used for solving or proposing alternative solutions to many problems in image analysis. We present a deep neural network based method for recovering the intravoxel structure from signals corresponding to diffusion weighted magnetic resonance images (DWMRI). Our neural network, with around 1 million of parameters, trained for recovering the compartment sizes of basis diffusion functions dictionary. We show results of our method on the analysis of synthetic data generated with the multi-tensor model (Tuch et al 2009) that are positively compared with those results computed with solution computed using NNLS with Lasso regularization for the same dictionary of basis functions.