Using Dynamic Time Warping and Agglomerative Clustering of ECG Data to Group Distinct PQRST Morphologies in Patients with Chagas Disease

Chagas disease is a significant public health issue in Latin America, affecting millions at risk of severe cardiac abnormalities.We aimed to capture distinct clusters of PQRST morphologies in individuals diagnosed with Chagas cardiomyopathy using unsupervised learning techniques.From 207 Holter recordings, we derived representative average heartbeats through ECG segmentation and K-Means clustering as pre-processing steps.We then computed a Dynamic Time Warping (DTW) distance matrix to perform agglomerative hierarchical clustering, assessing various cluster numbers (K) against silhouette scores for three linkage methods.Using complete linkage and K = 19 yielded a silhouette score of 0.182.Visual inspection confirmed separation of PQRST morphologies, capturing relevant patterns-including prevalent T-wave inversions, acute T-waves, and net positive or negative QRS complexes.We computed clinical metrics within each cluster to explore potential links between ECG morphologies and patient health.Averaging heartbeats made the process feasible due to the quadratic complexity of DTW.This study highlights the potential of this clustering approach in stratifying ECG data for high-level analysis, segregating ECGs with notable noise levels, and providing a tool for rapid morphological categorization.