Abstract

Convolutional Neural Networks (CNNs) have emerged as powerful tools for medical imaging analysis, which is essential for effective clinical interventions. However, standard CNNs often suffer from overfitting and sensitivity to noise in complex medical datasets, compromising diagnostic accuracy. While metaheuristic optimization techniques like Genetic Algorithms (GAs) can address these limitations, but struggles with slow convergence to achieve global optima solution. Over the years, improvement over GA has produced Elitist Genetic Algorithm (EGA) which is known as an efficient metaheuristic for complex structural optimizations, it improves upon GA by preserving the optimal solutions across generations, yet it remains highly susceptible to premature convergence and over-exploitation of restricted regions of the search spaces. Hence, this research developed an Adaptive and Diversity Genetic Algorithm (ADGA) by integrating adaptive parameter control to balance exploration and exploitation, alongside a diversity preservation technique to prevent premature convergence and used to optimize CNN which produces ADGA+CNN for better detection of heart conditions in Cardiac Magnetic Resonance Images (CMRIs). Cardiac Magnetic Resonance Images was acquired from Kaggle and augmented via left-right flipping. Preprocessing was performed using contrast enhancement, followed by segmentation using Fuzzy C-Means segmentation technique. Adaptive and Diversity Genetic Algorithm (ADGA) was formulated using adaptive parameter control and diversity preservation technique. The formulated ADGA was used to optimize CNN to achieve ADGA+CNN and translate to feature extraction. The performance of the ADGA+CNN was evaluated and compared with EGA+CNN and standard CNN based on False Positive Rate, Specificity Sensitivity, Precision, Recognition Accuracy and Computational Time. The empirical results demonstrate that the developed ADGA+CNN model significantly outperforms conventional CNNs and EGA-optimized model. The ADGA+CNN represents a powerful, highly automated, and reliable diagnostic tool that holds great potential to reduce human error, alleviate the workload of clinicians, and ultimately contribute to earlier and more accurate heart attack diagnoses, improving patient survival rates.