Quantitative Myocardial MR Perfusion: Accurate Delay Estimation through Deep-Learning based Outlier Detection

Motivation: While extensive research has explored robust pixel-wise quantification of myocardial blood flow, there is also a need for further investigation into accurate myocardial signal delay estimation, given its diagnostic value, for enhancing conventional clinical myocardial perfusion protocols. Goal(s): We seek to address the primary challenges in calculating delay and improve its estimation accuracy. Approach: We introduce a deep learning approach designed to recognize motion artifacts as outliers along the temporal signal curve of each voxel, subsequently eliminating them from the perfusion quantification process. Results: Our findings suggest that eliminating outliers enhances the accuracy of perfusion delay parameter estimation in scenarios with residual motion. Impact: Enhancing the precision of delay estimation will increase its diagnostic value as a biomarker, offering crucial insights into the identification of perfusion defects in ischemic hearts.