Bayesian inference using hierarchical and spatial priors for intravoxel incoherent motion MR imaging in the brain: Analysis of cancer and acute stroke

The intravoxel incoherent motion (IVIM) model allows to map diffusion (D) and perfusion-related parameters (F D*). Parameter estimation is, however, error-prone due the non-linearity of signal model, limited signal-to-noise ratio (SNR) small volume fraction perfusion in in-vivo brain. In present work, performance Bayesian inference was examined presence brain pathologies characterized by hypo- hyperperfusion. particular, a hierarchical spatial prior were combined. Performance compared relative conventional segmented least squares regression, only (non-segmented data likelihoods) deep learning approach. Realistic numerical IVIM simulations conducted assess errors ground truth. In-vivo, 11 central nervous system cancer patients 9 with acute stroke acquired. proposed method yielded reduced error for both scenarios other methods across whole investigated SNR range. contrast-to-noise better or on par techniques in-vivo. approach hence improves parameter stroke.