Deep Learning based Automatic Heart Disease Detection using ECG Signals

Abstract

For medical and clinical applications automated electrocardiogram (ECG) diagnosis may be very useful aid. We had implemented a deep learning approach in building a system for automatic classification and detection of ECG signals for processing. To detect cardiovascular disease in ECG signals we acquired expertise in convolutional neural network (CNN) by using a training data set of 259,789 ECG signals accumulated from cardiac function rooms in tertiary care hospital with facilities. Database provided availability of more than 4000 ECG signal samples taken from various outpatient ECG examinations gathered through 47 subjects: 22 females and 25 males. For normal class confusion matrix processed out from testing dataset showed 99% accuracy. In “atrial premature beat” class, ECG samples were accurately grouped 100% of time. Lastly, for “premature ventricular contraction” class, ECG segments was correctly segregated 96% of time. Totally, we found an average accuracy in classification of about 98.33%. Specificity (SPC) and sensitivity (SNS) was found to be 98.35% and 98.33% respectively.  A novel concept dependent on deep learning and, particularly on a CNN network ensuresoutstandingbehaviour in automated recognition hence helping in prevention of cardiovascular diseases and in some cases pre detection so as to take necessary preventive steps.