Cross-contrast Fusion and Aggregation Network for Multi-contrast MRI Super-resolution

Yue Ding, Tao Zhou, Ye Wu

Abstract: Magnetic resonance imaging (MRI) super-resolution (SR) can restore high-quality high-resolution (HR) images from low-resolution (LR) ones, which is helpful for clinical diagnosis and treatment. Different from SR restoration using a single-contrast, multi-contrast SR restoration has received increasing attention because different modalities can provide complementary information to improve the SR quality. However, fusing multi-contrast features as well as capturing cross-level correlations among different layers are still challenging. In this paper, we propose a Cross-contrast Fusion and Aggregation Network (CFA-Net) for multi-contrast MRI super-resolution, which can effectively explore complementary information from multi-contrast images to improve target-image SR quality. Specifically, the multi-contrast images are passed through multiple residual Swin Transformer blocks (RSTB) to learn hierarchical feature representations at different layers. Then, we present a Cross-contrast Fusion Module (CFM) to fuse the cross-contrast features in a layer-wise strategy, which can capture the complementary information from multi-contrast MR images. Moreover, a cross-level Feature Aggregation Module (FAM) is proposed to integrate cross-level features from CFMs for exploring the interaction between different layers. Experimental results on three multi-contrast MR datasets demonstrate that our method performs better than other state-of-the-art SR methods.