A Practical Algorithm for Improving Localization and Quantification of Left Ventricular Scar.

Current approaches to classification of left ventricular scar rely on manual segmentation of myocardial borders and manual classification of scar tissue. In this paper, we propose an novel, semi-automatic approach to segment the left ventricular wall and classify scar tissue using a combination of modern image processing techniques. We obtained high-resolution magnetic resonance angiograms (MRA) and late-gadolinium enhanced magnetic resonance imaging (LGE-MRI) in 14 patients who had ventricular scar from a prior myocardial infarction. We applied (1) a level set-based segmentation approach using a combination of the MRA and LGE-MRI to segment the myocardium and then (2) an automated signal intensity algorithm (Otsu thresholding) to identify ventricular scar tissue. We compared results from both steps to those of expert observers. The LVgeometry using the semi-automated segmentation method had a mean overlap of 94% with the manual segmentations. The scar volumes obtained with the Otsu method correlated with the expert observer scar volumes (Dice comparison coefficient of 0.85± 0.11). This proof of concept segmentation pipeline provides a more objective method for identifying scar in the left ventricle than manual approaches.