Automated myocardial scar segmentation on joint bright- and black-blood late gadolinium enhancement images

Motivation: Myocardial scar assessment is crucial for accurate patient prognostic. Magnetic resonance bright-blood sequences are used to retrieve heart anatomy information, while black-blood technique has shown promise for enhanced scar detection. However, manual delineation of infarction is time-consuming, operator-dependent and labor-intensive, limiting its accuracy and increasing its variability. Goal(s): To develop a fast, accurate and automated method for improved scar segmentation and analysis on co-registered bright- and black-blood images. Approach: Leverage an artificial intelligence-based pipeline to achieve automated scar segmentation. Results: Automated scar segmentation was achieved in an accurate and fast fashion. Good agreement with conventional manual segmentation methods were found. Impact: The proposed artificial intelligence-based pipeline exploits anatomical information and improved scar visualization from joint bright- and black-blood images. This permits fast, operator-independent, and time-saving scar segmentation and analysis for the radiologist, with required clinical quality to better help guide therapy.