Design and analysis of low power architecture for electrocardiogram abnormalities detection using artificial neural network classifiers

Many people around us are losing their life if they didn’t notice heart disease in the early stage and also ever-increasing medical cost for the treatment of cardiac-related health issues. The main research intention of this article is to design an efficient prototype system for abnormality detection using artificial neural networks on the Zed board FPGA platform. The datasets are taken for five abnormalities based on the heart rate variability (HRV). Artifacts are removed using the DCT algorithm and signal compression and feature extraction adaptive sensing advanced wavelets like symlet or biorthogonal. The Proposed design consists of artefact removal, abnormalities detection based on features extraction, classification of different abnormalities, and storage of ECG along with the type of abnormalities in the cloud. The artifacts removal can be done in two phases, firstly the ECG signals can be captured and the unwanted artefact can be removed after features can be extracted using DWT. Once the feature extraction has been completed, the abnormalities in the signal can be detected and identified by using MultiMate SVM classifiers. The design is successfully validated and verified for standard ECG signals datasets on FPGA, from the obtained results, it is found that 32% improvement in SNR, 14% in MSE, and 65% improvement in detection of abnormalities and also an accuracy achieved is 93.75% accuracy for the normal sinus rhythm abnormality.