ECG diagnostic support system (EDSS): A deep learning neural network based classification system for detecting ECG abnormal rhythms from a low-powered wearable biosensors

The latest developments in deep learning have made it possible to implement automated, advanced extraction of several things' features and classifications. Deep learning methods have also become more prominent in arrhythmia detection. This study conceptualized a classification method for ECG arrhythmia utilizing the Convolutional Neural Network (CNN) with images based on spectrograms without undergoing ECG visual examination such as R-peak or P-peak identification. This paper's CNN model would immediately disregard the noise parameter when its ECG data is converted into a 2D image while extracting the appropriate characteristic map in the pooling layer and convolution. Google's Inception V3 model was used to retrain the final layer of CNN for datasets recognition. This study established and formulated a diagnostic support system that enables the acquisition, interpretation, and analysis of clinical data and ECG biosignals from patients to facilitate heart disease diagnosis in rural areas or places where there is no ECG facility. Two ways were developed in training and testing the ECG datasets, the binary, and quinary classifications. These two classifications made a remarkable accuracy of 98.73% for binary and 97.33% for quinary. This study obtained a higher accuracy rate compared to the previous works. Specificity, sensitivity. Positive predictive values and F1 scores also made desirable results from 96.83% to 99.21%. Hence, we concluded that our system is an effective method in classifying heart rhythms to help the cardiologists in diagnostic analysis in the patient.