Comprehensive Summary
This multicenter, peer-reviewed study focused on determining whether an AI-assisted approach can help improve detection and quantification of myocardial scar compared to standard LGE imaging. The authors present SPOT, a multi-spectral imaging sequence that combines bright-blood and black-blood imaging in a single scan, paired with automated scar segmentation, using retrospective LGE cardiac MRI data from approximately 450 patients to evaluate ischemic and non-ischemic cardiomyopathy across multiple sites and scanners. The role of SPOT was to automate myocardial and ventricular scar segmentation, which were identified by deep learning–based computer vision models comprising convolutional neural networks with transformer elements. Performance was compared against that of conventional bright-blood PSIR imaging with expert manual segmentation as the clinical reference standard.
Outcomes and Implications
SPOT imaging demonstrated improved scar-to-blood contrast compared with phase-sensitive inversion recovery and identified a higher number of scarred myocardial segments, particularly at the subendocardial border, an area where conventional LGE does not perform consistently. Automated segmentation was highly consistent with expert readers and reduced interobserver variability. In clinical practice, this approach may improve consistency and efficiency of myocardial scar assessment.