Brain Tumor Classification from MRI Using Transfer Learning: A Comparative Literature Analysis and a MobileNetV2 Baseline with Preliminary Grad-CAM Analysis
- 1Netaji Subhas University of Technology, New Delhi, India
Abstract
Automated classification of brain tumors from magnetic resonance imaging (MRI) is an active area of research, but comparisons between published methods are challenging because of differences in datasets, classification tasks, and evaluation protocols. This study reviews five approaches to brain tumor analysis and presents a baseline classifier using MobileNetV2 transfer learning.
The baseline uses a frozen ImageNet-pretrained MobileNetV2 backbone and was trained for eight epochs on 7,200 MRI images across four balanced classes: glioma, meningioma, no tumor, and pituitary tumor. On a stratified held-out set of 1,440 images, the model achieved 93.3% accuracy and a macro F1-score of 0.93. Most classification errors involved confusion between glioma and meningioma.
The study also discusses limitations affecting the reliability and interpretation of the results, including the possibility of near-duplicate images across data partitions, the use of the held-out test set for monitoring during training, and the limitations of the preliminary Grad-CAM analysis. The findings position MobileNetV2 transfer learning as a useful baseline rather than a state-of-the-art solution and identify improvements needed for more rigorous evaluation and explainability.
Keywords: Brain tumor classification, MRI, deep learning, transfer learning, MobileNetV2, Grad-CAM, literature review.
Related
- Multi-crop disease classification using a Dilateformer-MobileNetV2 dual-branch fusion network
- Clinical Decision Support for Adult Enteral Nutrition: Literature Overview and Preliminary Hospital‐Based Evaluation
- Effect of p16 Immunohistochemistry on Interobserver Agreement in Cervical Biopsy Histopathology: A Structured Literature Review - Preliminary Synthesis
- Deep learning-based segmentation, grading and genotype classification of glioma: a literature analysis
- Transfer Learning-Based Classification of Alzheimer's Disease Stages from Brain MRI Using EfficientNetB0
BibTeX
@misc{agarwal2026brain,
title = {{Brain Tumor Classification from MRI Using Transfer Learning: A Comparative Literature Analysis and a MobileNetV2 Baseline with Preliminary Grad-CAM Analysis}},
author = {Ashna Agarwal},
howpublished = {Zenodo},
year = {2026},
month = oct,
doi = {10.5281/zenodo.23647318},
url = {https://zenodo.org/records/23647318},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Brain Tumor Classification from MRI Using Transfer Learning: A Comparative Literature Analysis and a MobileNetV2 Baseline with Preliminary Grad-CAM Analysis
AU - Ashna Agarwal
PY - 2026
DA - 2026/10/25
PB - Zenodo
DO - 10.5281/zenodo.23647318
UR - https://zenodo.org/records/23647318
AB - Automated classification of brain tumors from magnetic resonance imaging (MRI) is an active area of research, but comparisons between published methods are challenging because of differences in datasets, classification tasks, and evaluation protocols. This study reviews five approaches to brain tumor analysis and presents a baseline classifier using MobileNetV2 transfer learning. The baseline uses a frozen ImageNet-pretrained MobileNetV2 backbone and was trained for eight epochs on 7,200 MRI images across four balanced classes: glioma, meningioma, no tumor, and pituitary tumor. On a stratified held-out set of 1,440 images, the model achieved 93.3% accuracy and a macro F1-score of 0.93. Most classification errors involved confusion between glioma and meningioma. The study also discusses limitations affecting the reliability and interpretation of the results, including the possibility of near-duplicate images across data partitions, the use of the held-out test set for monitoring during training, and the limitations of the preliminary Grad-CAM analysis. The findings position MobileNetV2 transfer learning as a useful baseline rather than a state-of-the-art solution and identify improvements needed for more rigorous evaluation and explainability. Keywords: Brain tumor classification, MRI, deep learning, transfer learning, MobileNetV2, Grad-CAM, literature review.
ER -