Deep Feature Representation Learning For MRI-Based Brain Tumor Diagnosis Using Transfer Learning

Abstract

This paper presents a transfer learning-based approach utilizing the VGG16 architecture for the automatic detection and classification of multi-class brain tumors from magnetic resonance imaging (MRI) scans. The proposed method includes a preprocessing pipeline involving denoising, grayscale conversion, normalization, and resizing to 224 × 224 × 3. The VGG16 model is fine-tuned with dropout regularization and the Adam optimization algorithm. Experimental evaluation on a dataset of 1,343 MRI images comprising pituitary tumor, meningioma, glioma, and non-tumor classes achieved an overall accuracy of 94%, with average precision, recall, and F1-score of 0.94, 0.93, and 0.93, respectively. The results demonstrate that the proposed model provides robust and reliable performance, supporting its applicability as a clinical decision-support system for brain tumor diagnosis.