Design of an autocoder based secure system for authentication and arrhythmia detection with ECG signals

Electrocardiography (ECG) is a method used to measure the electrical activity of the heart, which varies between individuals. The fact that each person's heart structure and electrical characteristics are unique makes ECG signals applicable not only for medical diagnosis but also for biometric authentication systems. This study aims to develop reliable and innovative solutions for both the early detection of heart diseases and identity verification processes by analyzing ECG signals using deep learning techniques. In this study, arrhythmia detection and authentication tasks were carried out on ECG signals using the MIT-BIH Arrhythmia Dataset through deep neural network architectures. During the feature extraction phase, both a Spindle autoencoder architecture and a modified autoencoder architecture were employed. To ensure data security and protect personal privacy, the ECG signals were first compressed and then encrypted before processing. These encrypted signals were then trained using deep learning-based models for both arrhythmia detection and identity verification tasks. The results demonstrate that high accuracy rates were achieved in both cardiac disease diagnosis and authentication applications. As a result of the experiments, accuracy rates of 98.46% and 98.60% were obtained using two different methods for the authentication task. In the arrhythmia detection task, accuracies of 99.7% and 98.77% were achieved. These findings indicate that ECG signals possess significant potential as a biosignal for both medical diagnosis and biometric authentication.