Improving SASHA T1 precision via a deep-learning approach

SASHA with a 3-parameter fitting model has high T1 accuracy but low precision due to low SNR in saturation-recovery T1-weighted images. Alternatively, a two-parameter model could improve the precision with the penalty of losing accuracy. In this study, we developed a 1D neural network (DeepFittingNet) to predict SASHA T1 and alleviate the impaction from noise. We trained DeepFittingNet using simulation of SASHA with different Rician noise levels and tested it using in-vivo MR data. Results showed that T1 from DeepFittingNet had high precision and comparable accuracy to that of the 3-parameter fitting.