Gisela Vanesa Clemente, Leandro Andrini, Mariano Llamedo
Wavelet-based entropy and statistical complexity can quantify subtle alterations in the multiscale organization of ventricular depolarization that are not necessarily apparent during standard electrocardiographic inspection. This study evaluates the transferability, robustness, and methodological sensitivity of these descriptors in a large two-database post-myocardial-infarction setting. We analyzed 12-lead ECG records from the PTB Diagnostic Electrocardiogram Database and PTB-XL. A predefined analysis cohort contained 5,887 control (CTRL), 118 early/healing post-infarction (MI7), and 294 healed/chronic post-infarction (MI60) records. The supplied QRS matrices already contained segmented and aligned QRS windows. Because beat-level rejection metadata were not retained in these stored matrices, feature calculation was based on the QRS complexes available for analysis. Normalized wavelet entropy and statistical complexity were computed from relative wavelet energy and summarized by lead and by a multi-lead criterion. Statistical analyses accounted for repeated patients, age, sex, database origin, class imbalance, and patient-level separation during validation. In patient-grouped nested validation, the joint entropy-complexity representation yielded AUCs of 0.858 (95\% CI 0.808--0.902) for CTRL versus MI7, 0.621 (0.584--0.660) for CTRL versus MI60, and 0.759 (0.705--0.809) for MI7 versus MI60. Physically harmonized frequency analyses weakened several contrasts, particularly CTRL versus MI60. The descriptors remained associated with post-infarction group status after covariate adjustment, but their magnitude and discriminative performance depended partly on database composition and frequency definition. Wavelet entropy and statistical complexity are therefore best interpreted as exploratory, complementary signal-processing descriptors of multiscale QRS organization rather than as validated diagnostic or prognostic markers.