.png){kind=icon}
.png){kind=icon}
Return
Multi-domain ECG feature extraction enable accurate automated recognition of bipolar disorder, depression and schizophrenia
.png){kind=icon}
PRE
AI
X
B
Y
L
DOI：10.1016/j.jbi.2026.105026.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Objective: To develop and validate a multi-lead electrocardiogram (ECG)-based machine learning system for automated classification of major psychiatric disorders (bipolar disorder, major depressive disorder, and schizophrenia) using cardiac autonomic biomarkers, and to identify the most discriminative electrophysiological features for differential psychiatric diagnosis. Methods: A total of 233 de-identified 12-lead ECG records were retrieved from publicly available repositories (PTB Diagnostic ECG Database and MIMIC-IV-ECG), comprising 198 records with confirmed psychiatric diagnoses (62 bipolar disorder, 17 major depressive disorder, 119 schizophrenia) and 35 healthy-control records. Standard twelve-lead ECG recordings (mean duration 598.7 +/- 45.2 s) were acquired under resting conditions as documented in the source databases. Features were extracted independently from all twelve leads, yielding 1248 initial features, subsequently reduced to 84 optimal features through recursive feature elimination with cross-validation. Model performance was evaluated using stratified 10-fold cross-validation with bootstrap confidence intervals (n = 1000 iterations). Results: The ensemble model achieved overall diagnostic accuracy of 94.8% (95% CI: 92.1-96.8%) in distinguishing psychiatric disorders from healthy controls, with condition-specific sensitivities of 92.3% for bipolar disorder, 89.7% for major depressive disorder, and 95.1% for schizophrenia. The system demonstrated robust performance across age groups (93.8-95.2%) and maintained diagnostic accuracy above 85% at signalto-noise ratios down to 15 dB. Processing time averaged 12.7 +/- 2.3 s per participant on standard clinical workstations. Conclusion: This approach provides a non-invasive, cost-effective, and objective diagnostic modality with potential for integration into routine psychiatric assessment. Future research should further address medication confounding effects using larger multi-repository cohorts and evaluate longitudinal utility through prospective validation studies.
Keywords:
Psychiatric disorders
Schizophrenia
Autonomic nervous system
Electrocardiography
Journal
IF:
4.5
Papers:
3.5K
Citations:
1.9W
Organization
Cited Papers