Joint Reconstruction of Multi-contrast MR Images Through Information Decoupling, Alignment, and Fusion

Motivation: Multi-contrast MR images contain complementary information critical for quantitative MRI and disease diagnosis. Efficiently harnessing both contrast-variant and contrast-invariant information is vital for enhancing reconstruction quality and reducing acquisition times. Goal(s): We aim to develop a joint reconstruction model to efficiently reconstruct multi-contrast MR images. Approach: We propose a coarse-to-fine network architecture to effectively utilize inter-contrast information through decoupling, alignment, and fusion modules, while also leveraging intra-contrast information via multi-scale fusion. Results: Our model achieves a 1.39dB improvement in PSNR and a 2.48% increase in SSIM over the state-of-the-art methods for acceleration on an in-house multi-contrast dataset. Impact: Our joint reconstruction model significantly reduces acquisition times for multi-contrast MRI and shows promise for quantitative MRI, improving parametric map estimation. It also holds potential for other clinical applications requiring multiple imaging modalities.