Abstract
Compressed sensing (CS) in MRI accelerates data collection by undersampling in k-space and leveraging sparsity for precise reconstructions. Recent deep learning techniques excel in MRI reconstruction from such data, though many depend on intricate models. Newer deep reinforcement learning (DRL) methods offer quality image restoration by learning a policy for image refinement. However, DRL has limitations due to its limited action range and often neglects critical physics-based models seen in CS-MRI. Addressing this, we introduce an enhanced DRL framework that integrates model priors, improving DRL performance without extra computational burden. Tests show our approach outperforms previous standards.
References
C. Wang, R. Zhang, G. Maliakal, S. Ravishankar and B. Wen, "Deep Reinforcement Learning Based Unrolling Network for MRI Reconstruction," 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI), Cartagena, Colombia, 2023, pp. 1-5.