Abstract
Magnetic resonance imaging (MRI) is hard to categorize properly in terms of interclass similarity, there is data imbalance, and sensitive clinical decision-making: but the performance of convolutional neural networks (CNNs) highly relies on effective, yet computationally costly, hyperparameter tuning. To find solutions to such issues, the given paper proposes a hybrid solution to the problems of the Aquila Optimizer and Harris Hawks Optimization, i.e., Aquila Optimizer-Harris Hawks Optimization (AO-HHO) framework, to integrate the positive qualities of extremely good global exploration of the Aquila Optimizer and the good local exploitation process of a Harris Hawks Optimization to achieve balanced and robust CNN hyperparameter optimization. On a publicly accessible dataset of 7, 023 brain MRI images divided into glioma, meningioma, pituitary tumor, and non-tumor, the proposed algorithm has been tested on with fine-tuning critical hyperparameters, such as learning rate, batch size, number of filters, dropout rate, and optimizer type. The rate of accuracy, precision, recall and F1-score of the AO-HHO-tuned CNN is invariably high than the conventional metaheuristic algorithms, including the Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Whale Optimization Algorithm (WOA) that are approximately 78-83. The proposed method also helps in reducing the cost of computing. It takes only 77.85 s to train, while the baseline optimizers take more than 300 s. This shows that AO-HHO is a reliable, accurate, and computationally efficient framework that can be used for medical imaging decision-support applications that need to be done in real time and with limited resources.