Retrospective Detection and Suppression of Dark-Rim Artifacts in First-Pass Perfusion Cardiac MRI Enabled by Deep Learning.

The dark-rim artifact (DRA) remains an important challenge in the routine clinical use of first-pass perfusion (FPP) cardiac magnetic resonance imaging (cMRI). The DRA mimics the appearance of perfusion defects in the subendocardial wall and reduces the accuracy of diagnosis in patients with suspected ischemic heart disease. The main causes for DRA are known to be Gibbs ringing and bulk motion of the heart. The goal of this work is to propose a deep-learning-enabled automatic approach for the detection of motion-induced DRAs in FPP cMRI datasets. To this end, we propose a new algorithm that can detect the DRA in individual time frames by analyzing multiple reconstructions of the same time frame (k-space data) with varying temporal windows. In addition to DRA detection, our approach is also capable of suppressing the extent and severity of DRAs as a byproduct of the same reconstruction-analysis process. In this proof-of-concept study, our proposed method showed a good performance for automatic detection of subendocardial DRAs in stress perfusion cMRI studies of patients with suspected ischemic heart disease. To the best of our knowledge, this is the first approach that performs deep-learning-enabled detection and suppression of DRAs in cMRI.Clinical Relevance- Our approach enables clinicians to provide a more accurate diagnosis of ischemic heart disease by detecting and suppressing subendocardial dark-rim artifacts in first-pass perfusion cMRI datasets.