A Fine-Tuned Deep Transfer Learning-Based Framework for Efficient MRI Brain Tumor Images Classification

The expansion of brain tumor cells increases intracranial pressure within the skull, posing a serious threat to the delicate processes of life. Medication is the only solution to maximize lifetime. If not treated early on, it can be fatal. Thus, brain tumor categorization is a critical issue in the medical profession. Deep Transfer learning approaches (DTL) have demonstrated significant potential for brain tumor categorization. Utilizing magnetic resonance imaging (MRI) images of the brain skull. Transformer-based deep learning models have shown promising results for image categorization. This paper uses skull striping and mean filter to remove extra layers, denoise brain tumor cancer cell images, and classify using four pre-trained CNN architecture models: Alex Net, DenseNet-201, ResNet-152, and EfficientNet-B3. The approach employed in this study achieved the highest accuracy, with a value of 99.72%, using EfficientNet-B3. The ResNet-152 model also achieved a good accuracy rate of 99.23%. Other models have tumor classification accuracy values of DenseNet-201 98.62%, and Alex Net 98.22%, respectively. The findings indicate that EfficientNet-B3 has proven to be the most effective model for detecting brain tumor cancer cells. This level of accuracy facilitates early cancer detection, enabling identification before any physical symptoms appear.