Hybrid Aquila optimizer-Harris Hawks optimization for CNN hyperparameter tuning in brain tumor classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Magnetic resonance imaging (MRI) analysis faces challenges with interclass similarity, data imbalance, and sensitive clinical decisions.
• Convolutional neural networks (CNNs) performance heavily depends on hyperparameter tuning, which is often computationally expensive.
• Existing metaheuristic algorithms for CNN hyperparameter optimization may lack balance between exploration and exploitation.

Purpose of the Study:

• To propose a hybrid optimization framework, Aquila Optimizer-Harris Hawks Optimization (AO-HHO), for robust CNN hyperparameter tuning.
• To address the limitations of traditional optimization methods in medical imaging analysis.
• To enhance the accuracy and efficiency of brain tumor classification using MRI data.

Main Methods:

• Developed a hybrid AO-HHO framework integrating Aquila Optimizer's global exploration and Harris Hawks Optimization's local exploitation.
• Applied the AO-HHO framework to fine-tune critical CNN hyperparameters (learning rate, batch size, filters, dropout, optimizer type).
• Evaluated the framework on a dataset of 7,023 brain MRI images across glioma, meningioma, pituitary tumor, and non-tumor categories.

Main Results:

• The AO-HHO-tuned CNN achieved significantly higher accuracy, precision, recall, and F1-score compared to conventional algorithms (PSO, GA, WOA), with performance metrics ranging from 78-83%.
• The proposed AO-HHO framework demonstrated superior computational efficiency, reducing training time to 77.85 seconds compared to over 300 seconds for baseline optimizers.
• The framework effectively handled challenges like interclass similarity and data imbalance in brain MRI classification.

Conclusions:

• The AO-HHO framework offers a reliable, accurate, and computationally efficient solution for CNN hyperparameter optimization in medical imaging.
• This approach is suitable for real-time medical imaging decision-support applications with limited computational resources.
• The study highlights the potential of hybrid metaheuristic algorithms for advancing AI in healthcare.