Wavelet entropy and statistical complexity are best interpreted as exploratory, complementary signal-processing descriptors of multiscale QRS organization rather than as validated diagnostic or prognostic markers.
Key Findings
Results
The joint entropy-complexity representation achieved an AUC of 0.858 for distinguishing control records from early/healing post-myocardial infarction records.
AUC of 0.858 with 95% CI 0.808–0.902 for CTRL versus MI7 in patient-grouped nested validation.
MI7 group represented early/healing post-infarction records (n=118).
CTRL group contained 5,887 records.
Analysis used 12-lead ECG records from PTB Diagnostic Electrocardiogram Database and PTB-XL.
Statistical analyses accounted for repeated patients, age, sex, database origin, class imbalance, and patient-level separation during validation.
Results
Discriminative performance for distinguishing control records from healed/chronic post-myocardial infarction records was substantially lower than for early infarction.
AUC of 0.621 with 95% CI 0.584–0.660 for CTRL versus MI60 in patient-grouped nested validation.
MI60 group represented healed/chronic post-infarction records (n=294).
Physically harmonized frequency analyses further weakened several contrasts, particularly CTRL versus MI60.
This contrast was notably lower than the 0.858 AUC observed for CTRL versus MI7.
Results
The joint entropy-complexity representation achieved intermediate discriminative performance for distinguishing early/healing from healed/chronic post-infarction records.
AUC of 0.759 with 95% CI 0.705–0.809 for MI7 versus MI60.
MI7 (n=118) and MI60 (n=294) groups were drawn from both the PTB Diagnostic ECG Database and PTB-XL.
Performance fell between the CTRL vs. MI7 and CTRL vs. MI60 contrasts.
Results
Wavelet entropy and statistical complexity descriptors remained statistically associated with post-infarction group status after covariate adjustment.
Covariates included age, sex, database origin, and class imbalance.
Despite retained statistical association, the magnitude and discriminative performance of the descriptors depended partly on database composition and frequency definition.
Beat-level rejection metadata were not retained in the stored QRS matrices, which is a methodological limitation.
Results
Physically harmonized frequency analyses weakened the discriminative contrasts produced by wavelet entropy and statistical complexity, particularly for CTRL versus MI60.
The harmonized frequency analyses represented a methodological sensitivity test within the study design.
The CTRL versus MI60 contrast was specifically identified as most affected by this harmonization.
This finding indicates that frequency definition choices influence the descriptors' apparent discriminative ability.
Methods
The study analyzed a large two-database post-myocardial infarction cohort with marked class imbalance between control and infarction groups.
Total predefined analysis cohort contained 5,887 CTRL, 118 MI7, and 294 MI60 records.
Records were drawn from the PTB Diagnostic Electrocardiogram Database and PTB-XL.
Normalized wavelet entropy and statistical complexity were computed from relative wavelet energy and summarized by lead and by a multi-lead criterion.
QRS matrices were pre-segmented and aligned; beat-level rejection metadata were not retained.
Conclusions
The authors concluded that wavelet-based entropy and statistical complexity can quantify subtle alterations in multiscale ventricular depolarization organization not apparent on standard ECG inspection, but should be treated as exploratory rather than diagnostic markers.
Descriptors are described as capturing 'heartbeat micro-fragmentations' in the QRS complex.
The authors explicitly state these measures are 'best interpreted as exploratory, complementary signal-processing descriptors of multiscale QRS organization rather than as validated diagnostic or prognostic markers.'
Transferability and robustness were evaluated across two databases, and methodological sensitivity was assessed through frequency definition variations.
What This Means
This research suggests that two mathematical measures derived from heart rhythm recordings — wavelet entropy and statistical complexity — can detect subtle differences in the electrical patterns of the heart muscle (specifically in the QRS complex, which represents ventricular contraction) between people who have had heart attacks and those who have not. The study analyzed over 6,000 ECG records from two large databases, comparing healthy controls with patients in either the early healing phase or the chronic phase after a heart attack. The measures performed best at distinguishing early post-heart attack patients from healthy controls (accuracy measure of 0.858), but were less effective at distinguishing chronic post-heart attack patients from healthy controls (0.621), suggesting the electrical signal differences may be more detectable shortly after a heart attack than years later.
This research suggests that the way these measurements are calculated — particularly the frequency ranges used — significantly affects how well they work, and that results can vary depending on which database the ECG records come from. The study found that even after accounting for factors like age, sex, and database source, these measures still showed a statistically meaningful association with heart attack history, but their strength varied considerably based on methodological choices.
The practical implication of this work is that these wavelet-based signal measures could serve as additional, exploratory tools to help researchers study cardiac electrical changes after heart attacks, potentially picking up on subtle signal features that a clinician's eye would miss on a standard ECG. However, the authors are careful to note that these measures are not yet validated as standalone diagnostic or prognostic tools and should be considered complementary to, rather than replacements for, established clinical assessments.
Clemente G, Andrini L, Soria M. (2026). Quantifying heartbeat micro-fragmentations post-myocardial infarction using wavelet entropy and complexity.. Medical engineering & physics. https://doi.org/10.1088/1873-4030/aea228