AUTOMATED ARRHYTHMIA DETECTION USING ECG SIGNALS

Abstract

Arrhythmia of the heart is a life-threatening condition that prevents the heart from pumping enough blood to various organs of the human body. For many years, arrhythmia has been a major cause of death around the world, responsible for close to 12% of all deaths annually. When it comes to diagnosing these cardiac rhythm disorders, the Electrocardiogram (ECG) is the most accessible and cost-effective diagnostic tool. ECG is a non-invasive procedure that measures the electrical activity of the heart and provides enough information to interpret and classify cardiac rhythm disturbances. To diagnose arrhythmias using ECG records, cardiologists examine the rate and rhythm of the heartbeat. Despite being the most widely used diagnostic tool, machine-read ECGs are significantly inaccurate. As a result, classical algorithms cannot be used as a standalone diagnostic tool and are relegated to secondary roles. The situation is even direr in the middle- and low-income countries, where there is a dearth of heart specialists to diagnose abnormal rhythms. As a result, there is a need for an automated ECG analysis tool that is simple to use, accessible, reliable, and easy to interpret even by the nurses and paramedics. Keeping this in mind, this thesis uses single-lead ECG heartbeat signals for arrhythmia detection. Raw ECG signals are susceptible to noise, which can alter the ECG’s amplitude and time intervals, and these changes could be misdiagnosed as arrhythmias if appropriate filtering is not carried out. Filtering ECG signals is thus always a prerequisite to prevent false positives and incorrect diagnoses. As a result, adequate pre-processing is necessary before analyzing the ECG signal. Therefore, a novel neural network model named the attention-based convolutional denoising autoencoder (ACDAE) is proposed for denoising low-quality ECG signals. The proposed model is designed in such a way that the morphology of the signal is also restored while denoising the signal. The model utilizes a skip-layer and attention module for the reliable reconstruction of ECG signals from extreme noise conditions. A skip-layer connection reduces information loss while reconstructing the original signal. A lightweight, efficient channel attention module is used to update relevant features retrieved via cross-channel interaction efficiently. The model is trained and tested using four widely available standard databases. For evaluation, the signals are mixed with simulated additive white Gaussian noise ranging from −20 to 20 decibels (dB) and MIT-BIH Noise Stress Test Database (NSTDB) noise ranging from −6 to 24 dB. The model outperformed the most cited published works, achieving an average signal-to-noise ratio (SNR) improvement of 19.07±1.67 and a percentage-root-mean-square difference (PRD) of 11.0% at 0 dB SNR noise.