Related Experiment Videos
Global-local feature fusion: a robust hybrid deep learning model for multiclass brain tumor classification with
Kirti Pant1, Pijush Kanti Dutta Pramanik2,3, Shahid Mohammad Ganie4
1Department of Computer Science and Engineering, Bipin Tripathi Kumaon Institute of Technology, Dwarahat, Uttarakhand, India.
Frontiers in Genetics
|July 24, 2026
Summary
A novel hybrid Transfer Learning-Transformer model significantly improves brain tumor classification accuracy on MRI scans. This AI approach offers reliable and interpretable predictions for enhanced clinical decision-making.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Medicine
- Deep Learning for Diagnostics
Background:
- Accurate brain tumor classification from MRI is crucial but challenging due to manual assessment limitations.
- Existing deep learning models like CNNs and Transformers have limitations in capturing both local and global features.
- A hybrid approach is needed to leverage the strengths of both CNNs and Transformers for improved tumor classification.
Purpose of the Study:
- To develop a hybrid Transfer Learning-Transformer model for enhanced multi-class brain tumor classification.
- To integrate Convolutional Neural Network (CNN)-driven local feature extraction with Transformer-based global reasoning.
- To improve the accuracy and reliability of AI-assisted brain tumor diagnosis using MRI scans.
Main Methods:
- Benchmarked standalone CNN and Transformer models, including ResNet50, ViT, and others.
- Developed intra-family hybrid models (HTL, HTF) and a final fused hybrid model (HF).
- Validated performance on two public MRI datasets (Kaggle, Figshare) using metrics like accuracy, F1-score, AUC, and interpretability analysis (Grad-CAM++).
Main Results:
- The proposed hybrid model (HF) achieved near-perfect classification accuracy (up to 100%) on both datasets.
- HF significantly outperformed all baseline and single-architecture models.
- Statistical analyses confirmed HF as the top-performing model, with reliable probability estimation and tumor-focused decision-making.
Conclusions:
- The hybrid Transfer Learning-Transformer model demonstrates high accuracy and generalizability for brain tumor classification.
- The model provides reliable, interpretable predictions, making it a promising tool for AI-assisted diagnosis.
- This approach addresses limitations of standalone CNNs and Transformers, advancing medical image analysis.