Cardiac signal classification leveraging spectral optimization with ChebWave and deep blue particle filtering.

Abstract

Electrocardiogram (ECG) signal classification plays a critical role in diagnosing various cardiac conditions by identifying irregularities in heart rhythms. Despite advancements in the field, existing methodologies often rely on basic techniques that inadequately filter noise, leading to degraded performance and misinterpretation of vital features. This study presents the Spectral-Optimized Cardiac Framework (SOCF) approach to enhance the accuracy of ECG classification through advanced noise filtering, comprehensive feature extraction, efficient feature selection and integration of hybrid modelling techniques. The proposed methodology introduces the ChebWave Mean Refinement Filter (CWMRF) for effective noise reduction and to enhance signal clarity while preserving essential characteristics. In feature extraction, the Spectral Essence Extractor (SEE) captures both basic and high order features, providing deeper insights into ECG signals. Additionally, the Deep Blue Particle Optimizer (DBPO) efficiently identify relevant features while mitigating the risk of overfitting. Furthermore, the hybrid architecture of Convolutional neural network (CNN) and long short-term memory (LSTM) enable the model to effectively capture both spatial and temporal dependencies, thereby improving classification accuracy. To optimize performance, the Aquila Optimizer enhances the convergence speed and model efficiency by employing diverse search strategies inspired by the hunting behavior of Aquila bird. By integrating these advanced techniques, the SOCF achieved impressive results on the MIT-BIH dataset and PTB dataset with an accuracy of 99.6% and 99.68%, precision of 99.4% and 99.44%, recall of 99.5% and 99.51%, and F1 score of 99.2% and 99.49%, which significantly improves the robustness and reliability of ECG signal classification, ultimately providing more accurate clinical insights and better patient outcomes.