MGO-CRNN: a bio-inspired deep learning framework for accurate phonocardiogram classification

Abstract Accurate classification of phonocardiogram (PCG) signals is vital for the early detection of cardiovascular diseases (CVDs), a leading cause of global mortality. However, conventional machine learning and deep learning approaches often suffer from suboptimal feature extraction, class imbalance, and inefficient hyperparameter tuning. This study proposes an optimized hybrid CNN-RNN (CRNN) model, enhanced through the Mountain Gazelle Optimization (MGO) algorithm for automated hyperparameter tuning. The model incorporates onset peak detection for precise heartbeat localization, segmentation for structured analysis, and Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance. Evaluated on two benchmark datasets—PASCAL and PhysioNet2022—the MGO-CRNN model demonstrated substantial improvements in classification accuracy. Specifically, accuracy on the PASCAL dataset improved from 54.23% to 99.67%, and on PhysioNet 2022 from 99.50% to 99.99% after optimization. The optimized model significantly reduced classification errors and outperformed traditional tuning methods such as grid search and Bayesian optimization. These results underscore the effectiveness of MGO in enhancing model performance and generalization. The integration of advanced signal processing techniques and intelligent optimization provides a robust, real-time solution for AI-assisted cardiac screening and clinical decision support.