Determination of Image-based Biomarkers for the Diagnosis of Hypertrophic Cardiomyopathy, Hypertensive Cardiomyopathy and Amyloidosis From Texture Analysis in Cardiac MRI

摘要

Hypertrophic cardiomyopathy (HCM), hypertensive cardiomyopathy (HIP), and amyloidosis (AM) are pathologies in which a thickening of a portion of the myocardium occurs. All of them are manifested in a similar way on magnetic resonance images, which means that in most cases it is necessary to resort to the use of invasive diagnostic techniques. The objective of this work is to develop quantitative biomarkers that can differentiate between patients with these three pathologies using texture analysis on cardiac magnetic resonance imaging (MRI). In this study, a total of 103 patients underwent cine MRI. Two studies were carried out, one binary with patients with HCM and HIP and one multiclass considering the three pathologies. The left ventricular myocardium was segmented according to the standardized 17-segment model. A total of 43 features for each of the six segments were extracted using 5 different statistical methods. Four predictive models were implemented to evaluate the performance of the classification. Good precision results were obtained in both studies. For the binary study, a maximum AUC of $0.91pm 0.06$ was obtained with the K-Nearest Neighbours model and for the multiclass study the best performance (AUC $=0.89pm 0.12$) was achieved using the Support Vector Machine classifier.