Bayesian Inverse Regression for Vascular Magnetic Resonance Fingerprinting

Standard parameter estimation from vascular magnetic resonance fingerprinting (MRF) data is based on matching the MRF signals to their best counterparts in a grid of coupled simulated and parameters, referred as dictionary. To reach good accuracy, requires an informative dictionary whose cost, terms design, storage exploration, rapidly prohibitive for even moderate numbers parameters. In this work, we propose alternative dictionary-based statistical learning (DB-SL) approach made three steps: 1) quasi-random sampling strategy produce efficiently dictionary, 2) inverse regression model learn correspondence between fingerprints 3) use mapping provide both estimates confidence indices. The proposed DB-SL compared standard (DBM) method deep (DB-DL) method. Performance illustrated first synthetic including scalable with spatial undersampling noise. Then, are considered through simulations real acquired tumor bearing rats. Overall, two methods yield more accurate than range not limited boundaries. particular resists higher noise levels provides addition indices at no additional cost. appears promising reduce simulation needs computational requirements, while modeling sources uncertainty providing interpretable results.