Multi-class classification of brain tumors using optimized CNN and transfer learning techniques.

Brain tumors are abnormal cell growths within the brain or central nervous system that disrupt normal brain function and can be life-threatening. Early detection and precise classification of brain tumors are critical for effective treatment and improved patient prognosis. In this research, a Convolutional Neural Networks model with four convolution blocks is proposed for automated brain tumor detection and classification. The proposed scheme is applied to four types of MRI images, i.e., glioma tumor, meningioma tumor, no tumor, and pituitary tumor. The proposed architecture is optimized using the Adam optimizer, and the model is trained to minimize cross-entropy loss, enhancing classification accuracy across all four categories. The proposed model is compared with the three transfer learning models namely ResNet50, VGG19 and DenseNet121. Out of these models the proposed CNN model has outperformed with the value of accuracy as 98.5%, followed by second best accuracy at ResNet50 with value as 93.2%. After that, the category-wise analysis is performed on the proposed CNN model. Category 0 is numbered as Glioma, Category 1 as Meningioma, Category 2 as Pituitary, and Category 3 as No_tumor. Category 3 has obtained the highest value of accuracy as 0.98, followed by 0.96, 0.89 and 0.86 in category 2, category 1 and category 0 respectively. This analysis improves diagnostic accuracy for brain tumors and has the potential to aid radiologists in clinical settings.