Hypertrophic cardiomyopathy (HCM) is a condition characterized by an abnormal
thickening of the myocardium resulting in impaired blood pumping efficiency of the
heart. HCM exhibits clinical variability, with diverse symptoms and genetic backgrounds, resulting in pathophysiological complexity not fully captured by current
clinical guidelines.
Machine learning (ML) techniques have demonstrated utility in understanding and
classifying heart-related disorders, such as heart failure, as well as identifying ...
Hypertrophic cardiomyopathy (HCM) is a condition characterized by an abnormal
thickening of the myocardium resulting in impaired blood pumping efficiency of the
heart. HCM exhibits clinical variability, with diverse symptoms and genetic backgrounds, resulting in pathophysiological complexity not fully captured by current
clinical guidelines.
Machine learning (ML) techniques have demonstrated utility in understanding and
classifying heart-related disorders, such as heart failure, as well as identifying distinct
phenogroups within various conditions. This project aims to apply unsupervised
Multiple Kernel Learning (MKL) to characterize HCM patients based on electrocardiogram (ECG) and echocardiography measurements, contributing to a deeper
understanding of HCM’s underlying pathophysiological mechanisms and the identification of phenogroups.
More precisely, the project analyzes a single-center cohort comprising phenotypeand genotype-positive (G+) HCM patients (n = 73) and their phenotype-negative
relatives (n =32, 50% G+). All patients underwent digital 12-lead ECG, echocardiography, and a magnetic resonance (CMR) studies.
This analysis is based on the reduction of data dimensionality using unsupervised
MKL to combine the information from the input data to generate a space of reduced dimensions. Five different input data combinations are utilized: (1) using
only echocardiography measurements, (2) using only ECG complete cycles, (3) using only ECG segmented waves, (4) using echocardiography measurements and ECG
complete cycles and (5) using echocardiography measurements and ECG segmented
cycles. Regression to the input descriptors as a function of the newly computed
dimensions enables interpretation of how variance is represented in the new space.
Subsequently, clustering is performed in the resulting space, yielding four homogeneous patient groups that help address heterogeneity within the overall population
+