An effective arrhythmia classification via ECG signal subsampling and mutual information based subbands statistical features selection

The deployment of wireless health wearables is increasing in the framework of mobile health monitoring. The power and processing efficiencies with data compression are key aspects. To this end, an efficient automated arrhythmia recognition method is devised. The aim of this work is to contribute to the realisation of modern wireless electrocardiogram (ECG) gadgets. The proposed system utilizes an intelligent combination of subsampling, denoising and wavelet transform based subbands decomposition. Onward, the subband's statistical features are extracted and mutual information (MI) based dimension reduction is attained for an effective realization of the ECG wearable processing chain. The amount of information to be processed is reduced in a real-time manner by using subsampling. It brings a remarkable reduction in the proposed system's computational complexity compared to the fixed-rate counterparts. MI based features selection improves the classification performance in terms of precision and processing delay. Moreover, it enhances the compression gain and aptitudes an effective diminishing in the transmission activity. Experimental results show that the designed method attains a 4 times computational gain while assuring an appropriate quality of signal reconstruction. A 7.2-fold compression gain compared to conventional counterparts is also attained. The best classification accuracies of 97% and 99% are secured respectively for cases of 5-class and 4-class arrhythmia datasets. It shows that the suggested method realizes an effective recognition of arrhythmia.