Effective compression and classification of ECG arrhythmia by singular value decomposition

Abstract

Electrocardiogram (ECG) monitoring systems are widely applied to tele-cardiology healthcare programs nowadays, where ECG signals should always be compressed first during its transmission and storage. Previous studies attempted to achieve high quality decompressed signal with compression ratio as high as possible. In this paper, we investigated the performance on ECG arrhythmia classification on ECG signal decompressed after lossy compression with a high compression ratio. We proposed a simple but efficient method utilizing singular value decomposition (SVD) to decompose ECG signals, then applied the decompressed data to a convolutional neural network (CNN) and supporting vector machine (SVM) for classification. Using the optimization method with accuracy and compression ratio as objective functions, the highest average accuracy obtained is above 96% when the selected number of singular value is only 3. The evaluation results illustrated that the decompressed ECG signal even with a relatively high distortion can still achieve a satisfying performance in the arrhythmia classification.Thus,we proved that the real-time nature of the remote mobile ECG monitoring system can be greatly improved and countless people who are in need of ECG diagnosis can benefit from it.

Keywords

• ECG arrhythmia
• Supporting vector machine
• Convolutional neural network
• Singular value decomposition