Abstract: Heart beat classification is considered as the main tool for recognizing and diagnosing different heart diseases. The automation of heart beat classification is very necessary due to the exhaustive process of the 24 h mentoring of Electro Cardio Gram (ECG) signal of the heart. Moreover, ECG is considered as one of the most powerful tools for the diagnosis of heartbeats. In this study, a reliable automatic method is proposed separately on lead 1 and lead 2 to discriminate 15 classes of heart beat mapped into five main categories keeping into consideration the accuracy of each class besides the overall one. A dynamic segmentation strategy is applied to consider the heart rate variation. Discrete Wavelet Transform (DWT) is applied to extract beat features. Thereafter, the extracted features are subjected to Principle Component Analysis (PCA) to reduce the features dimension. Two different classifiers (Support Vector Machine (SVM) and random forests) are then applied on the reduced features to get the best results from the SVM classifier. Finally, the rejection method is applied to fuse the results from both leads 1 and 2. Using MIT-BIH as a validation database, SVM classifier achieved an overall accuracy of 99.5% and an average accuracy of 96.35% while random forests classifier achieved the best overall accuracy of 99.99% but an average accuracy of 84.26%. The study introduced also a comprehensive survey of recently researched work in the same application.
Hadeer El-Saadawy, Manal Tantawi, Howida A. Shedeed and Mohamed F. Tolba, 2017. Heart Disease Diagnosis using Electrocardiogram (ECG) Signal Processing. Asian Journal of Information Technology, 16: 771-782.