Applying Ventricular Wall Shape and Motion Features from CMRI for Aiding Diagnosis of Cardiomyopathies

Abstract

Cardiomyopathies are diseases usually characterized by dilation or hypertrophy of the heart muscle. Left Ventricle (LV) is the heart chamber most affected in most cases. Cardiac Cine Magnetic Resonance Imaging (CMRI) is a powerful tool applied for diagnosis of cardiomyopathies. Although some studies define descriptors based on CMRI images, usually they are related to clinical metrics. In this paper, we explored shape and motion features from the LV ventricular wall to define descriptors based on a priori knowledge about heart anomalies to build Supervised Machine Learning-based classification models capable of discriminating cases of dilated cardiomyopathy, hypertrophic cardiomyopathy, or those ones without anomalies associated with these diseases. The best classification model built and evaluated achieved F1-score = 0.85± 0.05, accuracy = 0.85± 0.04, and AUC = 0.94± 0.02. Our results are promising, indicating the potential of the approach for applications in computer-aided diagnosis systems.