EFFICIENT TECHNIQUE FOR ARRHYTHMIA DETECTION AND CLASSIFICATION USING ECG SIGNAL

Abstract

Cardiovascular health-related problem is a rapidly increasing integrated field concerning the processing and fetching the information from cardiovascular systems for early detection and treatment of cardiovascular diseases. Artificial Intelligence (AI) techniques, especially machine and deep learning techniques are more impactful and powerful tools for upgrading the capabilities of an application, and they have been applied to medical data for analysis and disease detection purposes. The work represents a comprehensive view of AI-based computational modelling with the abilities of powerful AI techniques that can play a crucial role in developing smart and enhanced systems in a real-world application. It consists of different techniques that are imposed over distinct ECG signals to evaluate the cardiovascular disease. It outlines the broad overview of AI-based modelling that can be utilized in various application domains. An electrocardiogram (ECG) plays a major role in biomedical applications to record the heartbeat activity. Regular monitoring of ECG through wearable devices like the band, watches, etc. can be done for early detection of cardiovascular diseases. The competency of each method discussed is related to ECG classification approaches that have been compared in terms of some parameters like accuracy, sensitivity, specificity, positive predictivity, and F1-score. The noise affects the ECG signal which may deteriorate the features of the respective signal that leads to improper treatment. De-noising has been done by pre- processing of the signal, which enables the prediction of the heart condition. Key morphological and statistical features are then extracted and used to train machine learning models for accurate classification of various arrhythmia types, such as atrial fibrillation, ventricular tachycardia, and premature contractions. The efficiency of ECG classification with different computational methods was evaluated with the executed algorithms by using different available databases. The challenges of existing techniques to analyse the ECG signal for the classification and detection of arrhythmia are summarized. The work demonstrates high accuracy, sensitivity, and specificity, highlighting its potential in automated cardiac monitoring systems. The findings vi underscore the effectiveness of integrating ECG signal analysis with intelligent classification models, offering a reliable tool for early detection and management of arrhythmias in clinical and remote healthcare settings. Here, arrhythmia detection and classification have been addressed through the integration of signal processing and time–frequency analysis techniques across different datasets. For segmentation and pre-processing of time-domain signal, a Group Sparse Mode Decomposition (GSMD) technique is employed to extract intrinsic mode functions, enabling detailed representation of ECG signals in terms of frequency and bandwidth. The noise and artefact present in ECG signal affect the signal in every aspect whether it is frequency, peak location, existence of ECG peaks etc. To eliminate the noise and artifacts present, a high resolution, Superlet Transform is applied, offering superior localization of transient features is necessary for identifying arrhythmic patterns. Additionally, Dynamic Mode Decomposition (DMD) is utilized to decompose ECG signals into distinct modes, capturing dynamic behaviors that support accurate arrhythmia classification. It is further applied to real- time ECG dataset to validate the model efficiency over arrhythmia detection. When integrated with machine learning classifiers, the features derived from DMD contribute to robust and accurate real-time arrhythmia classification, demonstrating its potential for practical deployment in clinical and wearable health monitoring systems. Together, these methods provide a comprehensive framework that significantly improves the detection and characterization of arrhythmias, contributing to more reliable and effective diagnostic tools. To ensure accurate localization of cardiac events, R-peak detection is carried out using Parallel Cluster Wavelet Analysis (PCWA). This method leverages the multi- resolution capability of wavelets along with unsupervised clustering to identify sharp variations in ECG signals that correspond to R-peaks. By operating across multiple frequency bands simultaneously, PCWA enhances the detection of peaks even in the presence of baseline wander, muscle artifacts, and noise.