Joint Tensor Low-Rank and Attention-based Sparse Unrolling Network for Accelerating Dynamic MRI

Motivation: Current unrolling networks that jointly utilize two priors in accelerating MRI adopted complex and nested iterative algorithm as their network structure foundation. Goal(s): To propose a simple yet efficient algorithm for unrolling network that jointly uses two priors. Approach: We propose a novel deep unrolling network, JotlasNet, for dynamic MRI reconstruction by jointly utilizing tensor low-rank and attention-based sparse priors. Based on the composite splitting algorithm, we design a simple yet efficient structure for the proposed JotlasNet. Results: Extensive experiments on a cardiac cine MRI dataset demonstrate the superior performance of JotlasNet in dynamic MRI reconstruction. Impact: The framework we proposed carries profound implications for various models incorporating joint priors, extending beyond the interaction of low-rank and sparse priors and transcending the realm of dynamic MRI reconstruction applications.