Improving quantitative MRI using self‐supervised deep learning with model reinforcement: Demonstration for rapid T1 mapping

• DOI: 10.1002/mrm.30045
• Publication Date: 2024-02-12T04:35:40Z
• Authors: Wanyu Bian; Albert Jang; Fang Liu
• Keywords: Robustness; Supervised Learning

Abstract Purpose This paper proposes a novel self‐supervised learning framework that uses model reinforcement, REference‐free LAtent map eXtraction with MOdel REinforcement (RELAX‐MORE), for accelerated quantitative MRI (qMRI) reconstruction. The proposed method an optimization algorithm to unroll iterative model‐based qMRI reconstruction into deep framework, enabling MR parameter maps are highly accurate and robust. Methods Unlike conventional methods which require large amounts of training data, RELAX‐MORE is subject‐specific can be trained on single‐subject data through learning, making it accessible practically applicable many studies. Using mapping as example, the was applied brain, knee phantom data. Results generates high‐quality correct image artifacts, removes noise, recovers features in regions imperfect conditions. Compared other state‐of‐the‐art methods, significantly improves efficiency, accuracy, robustness, generalizability rapid mapping. Conclusion work demonstrates feasibility new mapping, readily adaptable clinical translation qMRI.