Myocardial infarction detection from left ventricular shapes using a random forest

Understanding myocardial remodelling, and developing tools for its accurate quantification, is fundamental for improving the diagnosis and treatment of myocardial infarction patients. Conventional clinical metrics, such as blood pool volume or ejection fraction, are not always distinctive. Here we describe a method for the classification of myocardial infarction from 3D diastolic and systolic left ventricle shapes, represented by point sets. Classification features included global geometric, shape and thickness descriptors, and a random forest was used for classification. Results from cross validation show an accuracy of 92.5i¾?% leave-one-out and 91.5i¾?% 5-fold, improving the 87i¾?% obtained with ejection fraction thresholds. These results suggest that refined remodelling metrics provide information beyond standard clinical descriptors.