Improving Brain Tumor Classification from MRI Images Using EfficientNet

Brain tumors present a major health challenge, requiring precise and prompt detection to ensure effective treatment. This research focuses on preprocessing brain tumor MRI data and developing a robust classification model using the EfficientNet-B0 architecture. The study explores optimal configurations: employing the Adam optimizer with a learning rate 0.001, a batch size 32, and training over 12 epochs, with images resized to 224x224 pixels. During preprocessing, stringent techniques are applied to enhance image quality and feature extraction. These include resizing, mild sharpening filters, and brightness adjustment. Additionally, augmentation strategies are crucial, incorporating rotation, brightness, slight shifts in width and height, shear transformation, horizontal flipping, and using the nearest fill mode to enrich the dataset. The study utilizes an open dataset from Figshare comprising 3,064 MRI images, which are augmented to 12,256 images to increase dataset diversity and model robustness. Results demonstrate that the proposed model attains excellent performance metrics, achieving a testing accuracy of 99.91 %, a sensitivity (recall) of 99.8%, and a specificity of 99.96%. These findings highlight the effectiveness of the optimized EfficientNet-B0 model for brain tumor classification, emphasizing its potential for enhancing early diagnosis and treatment outcomes for patients.