Improving Brain Tumor Diagnosis using Convolutional Neural Networks and Transfer Learning

Tumors in the brain are lethal diseases in which proper and timely detection is essential for successful treatment. Classical diagnosis by expert-led appraisal of MRI images tends to be slow and subject to variability. This study presents an automated DL approach to classify tumors into four groups: glioma, meningioma, pituitary tumor, and non-tumor. Two strategies were examined: a custom convolutional neural network, transfer learning using MobileNet pre-trained on ImageNet. Preprocessing steps such as normalization, augmentation, and resizing were applied to improve dataset quality and model stability. Results indicate that MobileNet outperformed the custom CNN, offering higher accuracy and better generalization. These results illustrate the capability of light deep models to be effective and accurate tools for aiding radiologists in brain tumor classification.