Deep-learning-enabled Approach to Automatically Eliminate Motion-induced Dark-rim Artifacts in Stress First-pass Perfusion CMRI

Motivation: Dark-rim artifact (DRA) mimics perfusion defects in stress first-pass perfusion (FPP) cardiac MRI (CMRI), affecting the diagnostic accuracy in visual and quantitative methods. Goal: Eliminate motion-induced DRA in FPP CMRI. Goal(s): Eliminate motion-induced DRA in FPP CMRI. Approach: Apply temporal footprint (TF) reduction on scanner-reconstructed k-space to analyze effects on DRA severity in each myocardial sector (automatically segmented by a 2D+time deep neural network) and select the optimal TF reduction that minimizes DRA. Results: Experiments on truly healthy cases showed that proposed approach significantly reduces the severity and prevalence of DRA. We also showed the feasibility of inline implementation of proposed approach. Impact: Proposed approach offers an effective solution to eliminate DRA in first-pass perfusion CMRI. It helps improving the diagnostic accuracy of visual and quantitative methods, and the inline implementation can easily be deployed in routine clinical studies with standard CMRI protocols.