Automatic Detection System for the Classification of Tachycardias Using Intracardiac Electrograms

Abstract

The Electrophysiology (EP) study is a procedure performed in a catheterization lab to evaluate cardiac electrical activity and identify arrhythmias, aiming to predict potential abnormalities in cardiac rhythm. As an invasive method, the EP study records Intracardiac Electrogram (IEGM) and Electrocardiogram (ECG) signals simultaneously to aid in the diagnosis and treatment of arrhythmias. IEGM signals are obtained via catheters inserted within the heart, while ECG signals are captured through skin electrodes. During the EP procedure, electrophysiologists manually analyze and interpret these signals to identify specific tachycardias, an approach that is highly dependent on individual expertise and may vary in accuracy. To address these challenges, this research proposes an automated system to differentiate between various types of tachycardias, improving diagnostic consistency and efficiency. This approach not only assists EP specialists by minimizing manual effort but also has the potential to reduce the duration of EP procedures. The primary goal of this study is to develop a robust computational methodology for analyzing IEGM signals without compromising underlying signal activity, extracting relevant features across multiple domains for representation of the signal characteristics, and create automatic classification system capable of accurately differentiate different tachycardias. Initially, dataset of 66 patients was obtained from the Armed Forces Institute of Cardiology (AFIC) / National Institute of Heart Diseases (NIHD), Rawalpindi Pakistan. The data segments were then verified by the experienced electrophysiologists for different types of arrhythmias i.e., NSR, SVT, VT, AFL, AF and they were subsequently referred to as a gold standard for this research work. In general, the research was conducted across three stages. In the first stage, the obtained IEGM segments were preprocessed using several proposed computationally efficient algorithms to remove power line interference (PLI). The simulation results showed that the proposed (PDSSRLS and PD–SSLMSWAM) architectures provides almost the same qualitative and quantitative performances as those of the sequentially operated (SSRLS and SSLMSWAM) algorithms with less computational costs. In the second stage, feature extraction was conducted across multiple domains including time, frequency, time-frequency, and entropy in order to capture essential IEGM biomarkers. This includes techniques such as Nonlinear Energy Operator (NLEO) features (activity ratio, mean length active segment, standard deviation of active segments, mean length of inactive segments, Number of active segments), spectral estimation (including DF and APSR), entropy measures (Approximate Entropy and Sample Entropy) and wavelet-based features (RMS, iAV, mAV, WL, ZC, and SSC), resulting in a combined feature set of 15 distinct metrics. In the final stage, multistage classification framework was proposed to differentiate between different types of tachycardias and employed four classifiers in each stage to assess classification accuracy across feature combinations. Key classifiers include Decision Tree, Linear Discriminant Analysis, Support Vector Machine, and sk-Nearest Neighbors. Notably, the optimal performance achieved in this study combines all features with the DT classifier, yielding an overall classification accuracy and F–score of 90% and 96.5%, respectively. This methodology underscores the potential for automated systems to support electrophysiologists, offering improved accuracy and reduced procedural time in tachycardia diagnosis, thereby contributing significant advancements in clinical electrophysiology.