3D Hough Transform-Based Left Ventricle 3D Object Classification for Cardiomyopathy Diagnosis

Cardiomyopathies are diseases characterized by anomalies in the myocardium that in most cases mainly affect the left ventricle of the heart. The progression of these diseases can lead to heart failure and arrhythmias, as well as increase the risk of sudden death. Cardiac Magnetic Resonance Imaging (CMRI) is an important tool for the diagnosis of these diseases. However, a CMRI exam produces dozens of images over a period of time that need to be visually analyzed by the specialist to compose a diagnosis. This task is exhaustive and can often lead to visual fatigue, compromising the accuracy of the diagnosis. In this research, we investigated the application of a 3D Hough Transform-based feature descriptor, presented in a previous work, combined with Supervised Machine Learning algorithms to classify left ventricle 3D objects reconstructed from CMRI slices. In terms of $\mathrm{F}_{1}-\text{score}$ and accuracy, we observed overall mean classification performance of up to 0.75. In terms of AUC, the overall mean performance reached 0.89, indicating promising results with high potential of application.