Sudden cardiac death (SCD) represents a critical acute cardiovascular event characterized by rapid onset of cardiac and respiratory arrest, posing a significant threat to patients due to its high fatality rate. Monitoring indices related to SCD using wearable devices holds profound implications for preemptive measures aimed at reducing the incidence of such life-threatening events. Hence, this study proposed a predictive algorithm for SCD leveraging single-lead electrocardiogram (ECG) signals featuring low signal-to-noise ratios. Initially, simulated electrode motion artifact noise was introduced to ideal ECG signals to emulate the signal conditions with low signal-to-noise ratios encountered in everyday scenarios. To meet the criteria of simplicity and cost-effectiveness required for wearable devices, the analysis focused exclusively on single-lead signals. The proposed algorithm in this study employed a lightweight machine learning approach to extract 12-dimensional features encompassing ventricular late potentials, T-wave electrical alternation, and corrected QT intervals from the signal. The algorithm achieved an average prediction accuracy of 93.22% within 30 min prior to SCD onset, and 95.43% when utilizing a normal sinus rhythm database as a control, demonstrating robust performance. Additionally, a comprehensive Sudden Cardiac Death Index (SCDI) was devised to quantify the risk of SCD, formulated by integrating pivotal two-dimensional features contributing significantly to the algorithm. This index effectively distinguishes high-risk signals indicative of SCD from normal signals, thereby offering valuable supplementary insights in clinical settings.
Read full abstract