Diagnosis of Diseases by ECG Using Convolutional Neural Networks

The aim of the Master's degree dissertation was the implementation of an algorithm for the automatic arrhythmia classification using convolutional neural networks (CNN). The experiments showed that the use of neural networks in the task is possible and, moreover, is a promising choice. In the paper, a comparison of existing architectures was carried out, a new EcgResNet34 was designed, and a program that could automatically analyze the cardiogram and find abnormal heartbeats was developed. All the code is public, that allows researchers to get acquainted with the presented work and integrate automatic analysis of the electrocardiogram into their business or research product. The final result of the correct answers (accuracy) on the validation MIT-BIH arrhythmia database is 99.38%, which is the best result compared to other SOTA approaches.