Cardiac abnormality detection with a tiny diagonal state space model based on sequential liquid neural processing units

Abstract

This manuscript presents and studies the performance of the Diagonal State Space Sequence (S4D) model based on the Closed-form Continuous-time (CfC) network in order to achieve a high-performing cardiac abnormality detection method that is robust, generalizable, and tiny in size. Our S4D-CfC model is evaluated on 12- and 1-lead electrocar-diogram (ECG) data from over 20,000 patients. The system exhibits validation results with strong average F1 score and average AUROC value of 0.88 and 98%, respectively. To demonstrate the tiny machine learning (tinyML) of our 242 KB size model, we deployed the system on relatively resource-constrained hardware to evaluate its training performance on the edge. Such on-device fine-tuning can enhance personalized solutions in this context, allowing the system to learn each patient’s data features. A comparison with a structured 2D Convolutional LSTM (ConvLSTM2D) CfC model (ConvCfC) demonstrates the S4D-CfC model’s superior performance. The size of the proposed model is also significantly small (25 KB) while maintaining reasonable performance on 2.5s data, 75% shorter than the original 10s data, making it suitable for resource-constrained hardware and reducing latency. In summary, the S4D-CfC model represents a groundbreaking advancement in cardiac abnormality detection, offering robustness, generalization, and practicality with the potential for efficient deployment on limited-resource platforms, revolutionizing healthcare technology.