Title:Brain Tumor Classification Using Transfer Learning on Preprocessed MRI Images


Authors:M.Sathish, Bibin Graceson J, Ramanan K R


Published in: Volume 3 Issue 1 Jan June 2026, Page No.426-433


DOI: 10.63844/IJAITR.v3.i1.2026.426-433 cite


Keywords:Brain Tumor Detection, MRI, U-Net, Deep Learning, Image Segmentation, Morphological Operations, Transfer Learning, CNN Encoders, Preprocessing, Post-Segmentation, Classification


Abstract: Accurate and early detection of brain tumors is critical for successful treatment planning and patient outcomes. This project proposes an enhanced brain tumor detection framework that combines U-Net, a powerful deep learning-based image segmentation architecture, with traditional morphological operations for improved accuracy. Accurate and early brain tumor detection is crucial for treatment and patient outcomes The existing framework utilizes a U-Net for pixel-level tumor segmentation on preprocessed MRI images, refined by morphological operations to enhance accuracy and boundary definition. This system has demonstrated high segmentation reliability. Building on this, our proposed system significantly enhances the U-Net architecture by integrating transfer learning with pre-trained CNN encoders. This aims to profoundly improve segmentation accuracy by leveraging rich, hierarchical feature representations learned from vast, general image datasets. We will also explore more advanced preprocessing and adaptive post-segmentation strategies to further optimize the U- Net's performance. The enhanced U-Net will deliver superior pixel-level segmentation and contribute to a more precise diagnostic pipeline, evaluated for state-of-the-art accuracy in tumor segmentation and classification on diverse MRI datasets, encompassing various scanner types and patient demographics.


Download PDF