Robust Multi-Contrast MR Reconstruction Based on Disentangled Representation Learning-Embedded Deep Unrolling

Motivation: Multi-contrast MR imaging is common in clinical practice, creating a need for a universal reconstruction model that can effectively handle various contrasts in MRI. Goal(s): We aim to develop a reconstruction model that performs well on multiple contrasts. Approach: We developed a model that leverages disentangled representation learning to disentangle the MR image into contrast-specific style and contrast-invariant content representations during reconstruction. Results: The proposed model successfully generates disentangled style representations and aligned content representations, demonstrating superior performance compared to other methods in our reconstruction experiments. Impact: This DRL-assisted reconstruction approach has the potential to serve as a universal model for multi-contrast MR data.