Wearable ECG-Based Ventricular Fibrillation Detection Using Convolutional Neural Networks

Timely detection of ventricular fibrillation (VF) is essential in the fight for deadly cardiac arrhythmias. This research introduces a wearable electrocardiogram (ECG)-based VF diagnosis system using Convolutional Neural Networks (CNNs) for real-time evaluation. The proposed CNN model autonomously derives spatial and temporal patterns from raw ECG data, hence improving detection accuracy, in contrast to previous approaches that depend on created features. The model was trained and assessed using a publically accessible ECG dataset, attaining an accuracy of 98.5%, sensitivity of 97.8%, and specificity of 99.1% in differentiating VF from normal and other arrhythmic states. Data augmentation methods and refined hyperparameters enhanced generalization across various ECG signal variances. Further validation was conducted on the system utilising wearable ECG data collected in real-time, with an average inference time of 12 ms per sample, enabling quick identification appropriate for continuous monitoring applications. The proposed CNN-based method markedly improves traditional rule-based and machine-learning methods regarding performance and computing efficiency. Incorporating this model into wearable health monitoring devices may enable early diagnosis of VF, decrease reaction time for medical treatment, and improve the results for patients. Future efforts will focus on practical implementation and improving power efficiency for edge computing.