A Low-Cost and stable DAL arrhythmia detection algorithm based on the weak stratification query strategy of morphological statistical features

Abstract

The annotation cost of electrocardiogram (ECG) data is extremely high, resulting in a lack of labeled data. However, most existing models are based on supervised learning and highly dependent on labeled data. Therefore, this study proposes a low-cost and stable arrhythmia detection algorithm based on deep active learning to reduce annotation costs and develop a model with a low dependence on labeled data. The algorithm first proposes a Skew series query strategy based on the weak stratification of morphological statistical features, especially for ECG data, including Skew, Skewierste, Skewier-C, and Skewier-C/2, and develops a CNN-based classifier. Finally, experiments verified that the query strategy proposed in this sstudy has higher stability and adaptability than other classical ECG strategies, and that the performance of the proposed CNN is also higher than that of other classical classifiers. The heartbeat detection algorithm based on deep active learning (DAL) proposed in this study can significantly reduce the dependence on labeled data, significantly reducing annotation costs.