Machine Learning for Quantitative IVIM (qIVIM) Cerebral Perfusion Imaging

Intravoxel Incoherent Motion (IVIM) is a non-contrast perfusion scan that measures various speeds of molecular movement. Segmentation of IVIM signal origin, i.e. from blood, cerebrospinal fluid (CSF), or tissue water, is crucial. Simulation suggested Inversion Recovery should not be used in IVIM to remove CSF, and analysis of signal patterns suggested differing signal decay as a method of removing CSF during post-processing. A threshold determined by k-fold leave-one-out cross-validation on IVIM diffusion coefficient returned successful CSF segmentation (dice = .69), and CSF removal by supervised machine learning via linear discriminant analysis returned quantitative IVIM with strong agreement to microsphere perfusion.