Abstract

Convolutional Neural Networks (CNNs) have shown significant promise in addressing the complex challenge of classifying MRI images for detecting Alzheimer’s disease and identifying brain tumors. Despite their ability to automatically optimize parameters during training, determining the optimal hyper-parameter values remains difficult due to the vast and intricate search space, often leading to suboptimal outcomes. Researchers frequently resort to trial-and-error or rely on expert knowledge to navigate this challenge, as the training process does not ensure globally optimal parameter settings. This limitation hinders the development of effective real-world applications for MRI image analysis. To overcome this, we propose a novel hybrid model that integrates Particle Swarm Optimization (PSO) with CNNs to improve classification performance. The PSO algorithm is employed to identify the optimal CNN hyper-parameter configuration, which is then applied to the CNN architecture for enhanced classification. Our approach achieves higher prediction accuracy and minimizes loss in function value for brain disease detection. We evaluated the model using three benchmark datasets: the Alzheimer’s Disease Neuroimaging Initiative (ADNI), a Kaggle international dataset for Alzheimer’s disease, and a brain tumor dataset. Experimental results highlight the model’s superior performance, with accuracy rates of 98.50%, 98.83%, and 97.12% for the respective datasets, demonstrating its effectiveness in MRI-based disease classification.