Advanced ECG Signal Analysis Using AM-RNN and Optimized Feature Selection for Improved Cardiac Diagnosis

Abstract

Electrocardiogram (ECG) signals are essential in diagnosis and monitoring of cardiac conditions. However, effective and accurate classification of these signals remain as a challenge, due to noisy nature of data and increased complexity in distinguishing between various cardiac patterns. This work presents novel approach for ECG signal classification through an integrated deep learning framework. Initially, ECG signal pre-processing is performed where normalization is done to mitigate noise and ensure consistency across the dataset. Following normalization, feature extraction is conducted using Power Spectral Density (PSD) analysis, which identifies the power distribution of signal in frequency domain. To further enhance classification performance, a feature selection process is implemented using the Monarch Butterfly Optimization (MBO) algorithm, which efficiently reduces the dimensionality of feature set by selecting the most relevant and discriminative features. The selected features are subsequently fed into an Attention-infused morphological Recurrent Neural Network (AM-RNN) for classification. The AM-RNN incorporates adaptable filters that adjust based on input data resulting in superior performance in classifying various cardiac conditions. The proposed system is executed using Python software, and the results demonstrate the superiority of the work in comparison to existing techniques.