Analyzing Clinical 12-Lead ECG Images Using Deep Learning Algorithms for Objective Detection of Cardiac Diseases

Electrocardiogram (ECG/EKG) is the most common method for the study and detection of cardiovascular diseases. Current clinical ECG devices generate 12-lead ECG traces as images on paper. The majority of artificial intelligence (AI) algorithms created for automated cardiac monitoring are based on ECG data, which necessitates brand-new, expensive equipment that the majority of clinics cannot afford. In this paper, we propose a novel method of using deep learning (DL) techniques to analyze clinical 12-lead ECG images for the objective detection of cardiac diseases. A convolutional neural network (CNN) is a DL technique that uses 2D images as input and convolves them with various filters to produce the required outputs. CNNs can be trained with enormous datasets and millions of parameters. This work introduces a high-performance CNN-based method for the objective diagnosis of heart disorders in ECG images. The proposed model automatically learns a suitable feature representation from raw clinical ECG images and thus negates the need for hand-crafted features. The ECG image dataset of 929 distinct patient records, which contains 12 lead ECG information of different cardiac patients from the Mendeley Database, was used to evaluate the classification performance. Before being analyzed by CNN, all clinical ECG waveform images were converted into two formats: colorful and grayscale images. The proposed system achieved a maximum of 97% accuracy and 96% sensitivity for colored images and 98% accuracy and 97% sensitivity for grayscale images. To validate the result, we classified the images by using dense artificial neural networks (dense ANN) and compared the results with our CNN results and CNN significantly improved the accuracy. As this proposed method is highly accurate and does not economically burden clinics, it can potentially be used as a clinical auxiliary diagnostic tool and effectively optimize medical resources.