Hybrid deep learning-based ecg image classification using an enhanced parrot optimization algorithm for segmentation and feature selection

Abstract

Abstract Cardiovascular diseases represent a significant global health burden, emphasizing the importance of precise and computationally efficient diagnostic approaches. In medical image analysis, determining optimal thresholds for multilevel segmentation is crucial for detecting cardiac disorders. Feature selection serves a vital role in reducing dimensionality, removing redundant information, and enhancing both accuracy and interpretability. Conventional feature selection techniques tend to lose effectiveness as data complexity increases, often yielding suboptimal outcomes. To overcome these limitations, a modified Parrot Optimization algorithm, designated POesq, was developed by incorporating an Enhanced Solution Quality mechanism that accelerates local search and mitigates premature convergence. Experimental evaluation was conducted in two phases. The first involved multilevel ECG image segmentation, where POesq was benchmarked against established optimization algorithms using PSNR, SSIM, FSIM, and MSE. Results, supported by a Friedman mean rank test, demonstrated consistently superior segmentation performance. The second phase utilized POesq for feature selection within a hybrid deep learning framework that combines MobileNet and InceptionV3, achieving 98.66% accuracy, 98.24% precision, 98.21% sensitivity, and 98.21% F1-score, thereby surpassing benchmark methods.