Speaker
Description
Biological systems, such as the cardiovascular system, are governed by complex nonlinear dynamics arising from interactions within and between their components. From the perspective of statistical physics, metrics from information theory provide a powerful framework for quantifying couplings among physiological processes. In particular, entropy-based methods enable the detection of directional information flows between physiological time series.
This study investigates the information flow between heart rhythm and repolarization processes in patients with Long QT Syndrome (LQTS). Data from 195 LQTS patients (including subtypes LQTS1, LQTS2, and LQTS3) and 150 healthy controls were selected from the THEW database, focusing on nighttime Holter ECG recordings. From these signals, RR, QT, and DI intervals, as well as QRS and T-wave amplitudes, were extracted. After segmentation and detrending, signal stationarity was verified. Conditional entropies were estimated in bivariate and trivariate configurations using the ITS Toolbox developed by Luca Faes and collaborators. Out of 23 entropy-based variables, only those representing statistically significant information flow were retained.
Classification models using Random Forest (RF) and Support Vector Machine (SVM) were trained on the selected features to distinguish LQTS patients from healthy individuals. For RF, both accuracy and specificity were above 90%, while for SVM, accuracy, sensitivity, and specificity all exceeded 90%. These findings demonstrate that entropy-based metrics derived from ECG recordings can effectively capture the complexity of cardiac electrophysiology and support the development of automated diagnostic tools for arrhythmogenic disorders such as LQTS.
