Uncertainty-Aware Self-supervised Neural Network for Liver T_1ρ Mapping with Relaxation Constraint
share
T_1ρ mapping is a promising quantitative MRI technique for the non-invasive assessment of tissue properties. Learning-based approaches can map T_1ρ from a reduced number of T_1ρ weighted images, but requires significant amounts of high quality training data. Moreover, existing methods do not provide the confidence level of the T_1ρ estimation. To address these problems, we proposed a self-supervised learning neural network that learns a T_1ρ mapping using the relaxation constraint in the learning process. Epistemic uncertainty and aleatoric uncertainty are modelled for the T_1ρ quantification network to provide a Bayesian confidence estimation of the T_1ρ mapping. The uncertainty estimation can also regularize the model to prevent it from learning imperfect data. We conducted experiments on T_1ρ data collected from 52 patients with non-alcoholic fatty liver disease. The results showed that our method outperformed the existing methods for T_1ρ quantification of the liver using as few as two T_1ρ-weighted images. Our uncertainty estimation provided a feasible way of modelling the confidence of the self-supervised learning based T_1ρ estimation, which is consistent with the reality in liver T_1ρ imaging.
READ FULL TEXT