LSTM-Based Classification of Psychiatric Disorders Using Ultra-Short Raw ECG Signals

Psychiatric disorders, including bipolar disorder, depression, and schizophrenia, pose substantial public health challenges due to the intricate nature of their diagnosis and management. Electrocardiogram (ECG) signals offer a noninvasive method to assess autonomic nervous system function, which is frequently dysregulated in these conditions. However, conventional ECG analysis techniques often require prolonged recordings and extensive preprocessing, limiting their applicability in real-time or large-scale settings. This study presents an approach leveraging Long Short-Term Memory (LSTM) networks for medical data classification, specifically for the identification of psychiatric disorders using ultra-short, raw ECG signals. Our methodology efficiently captures the temporal dynamics of ECG signals within brief windows, significantly reducing the need for extensive preprocessing and manual feature extraction. The proposed LSTM-based classifier achieved an impressive mean accuracy of 97.12% ± 0.006, underscoring its precision and efficacy in distinguishing between different psychiatric conditions. Our findings indicate that LSTM networks, with their capacity to effectively model complex temporal patterns in short ECG recordings, represent a promising advancement in the diagnostic landscape of psychiatry.