Parallel Multi-scale convolution based prototypical network for few-shot ECG beats classification

Abstract

The electrocardiogram (ECG) presents essential information of the electrical activity of the heart measured by electrodes placed on the body surface, forming an important approach to diagnosing cardiac arrhythmias. Although various deep-learning based models have been implemented for auto-classification of arrhythmias, limited clinical data still impedes the progress of auto-diagnosis. This study presented a parallel multi-scale convolution based prototypical network (PM-CNN ProtoNet) for processing the few-shot learning tasks of ECG beats classification. By evaluating the proposed model on the MIT-BIH arrhythmia database, the PM-CNN ProtoNet achieves a satisfying accuracy of 91.6% in a 2-way 10 shot task. The comparative results between the PM-CNN ProtoNet and other state-of-art models also demonstrate the efficiency of our proposed model. In conclusion, the PM-CNN structure can improve the classification performance of the prototypical network in few-shot learning tasks while having the potential for auto-classification under a small amount of medical data.