Uncertainty-Aware Self-supervised Neural Network for Liver T1ρ Mapping with Relaxation Constraint

T1ρ mapping is a promising quantitative MRI technique for the non-invasive assessment of tissue properties. Learning-based approaches can map T1ρ from a reduced number of T1ρ weighted images, but requires significant amounts of high quality training data. Moreover, existing methods do not provide the confidence level of the T1ρ estimation. To address these problems, we proposed a self-supervised learning neural network that learns a T1ρ mapping using the relaxation constraint in the learning process. Epistemic uncertainty and aleatoric uncertainty are modelled for the T1ρ quantification network to provide a Bayesian confidence estimation of the T1ρ mapping. The uncertainty estimation can also regularize the model to prevent it from learning imperfect data. We conducted experiments on T1ρ data collected from 52 patients with non-alcoholic fatty liver disease. The results showed that our method outperformed the existing methods for T1ρ quantification of the liver using as few as two T1ρ-weighted images. Our uncertainty estimation provided a feasible way of modelling the confidence of the self-supervised learning based T1ρ estimation, which is consistent with the reality in liver T1ρ imaging.