Brain Cancer MRI Classification using CNN Architectures with Grid Search Enhanced with Coordinate Ascent Optimization

Magnetic resonance imaging (MRI) is an important tool for brain cancer diagnosis and classification. Combined with modern convolutional neural network (CNN) technology, it can effectively improve the accuracy and efficiency of tumor classification and provide an important reference for clinicians. Previous studies have shown that CNN has advantages in medical image classification. However, indepth discussion of the performance differences and hyperparameter optimization of different CNN models in brain cancer MRI image classification remains insufficient, limiting model selection and accuracy improvements in clinical applications. This study aims to provide a CNN model selection and optimization methodology for brain cancer MRI image classification to improve classification accuracy and reliability. The study selected CNN models of different levels and complexities, including LeNet-5, AlexNet and ResNet-18, and used the Grid Search Enhanced with Coordinate Ascent (GSECA) method to optimize hyperparameters, providing a feasible model selection and optimization methodology for brain cancer MRI image classification. Experimental results show that the method described in this paper achieved excellent results when using the Bangladesh Brain Cancer MRI Image Four-Classification Dataset (PMRAM) for classification. Based on the classification of the ResNet-18 model, the test accuracy reached 95.69%, and the $F 1$-scores of the four categories of Glioma, Meningioma, Normal and Pituitary reached 94.33%, $94.07 \%, 96.12 \%$ and 98.21% respectively. The precision of Pituitary was as high as $\mathbf{9 9. 1 0 \%}$. This result was significantly better than AlexNet and LeNet-5, and was at an excellent level in similar studies, achieving effective model screening. At the same time, it verified the importance of network depth and effective hyperparameter optimization to improve the performance of classification tasks in this scenario.