An explainable hybrid transformer–CNN framework for multi-class brain tumor classification from MRI images
Abstract
The identification of brain tumors through magnetic resonance imaging (MRI) requires effective methods that provide precise results because these results determine clinical choices. The study presents a classification system based on hybrid Transformer-CNN technology which classifies brain tumors into multiple types by using tumor segmentation and deep feature extraction and explainable artificial intelligence. The preprocessing of MRI images involves several steps which include skull stripping and CLAHE-based contrast enhancement and normalization and data augmentation and resizing to create high-quality images that maintain uniformity for model input. The TransUNet architecture performs tumor region segmentation by integrating convolutional neural networks with transformer-based attention systems which enable it to recognize both nearby and distant contextual details. The system generates segmentation masks which enable region-of-interest (ROI) feature extraction through DenseNet121 that has been pretrained on ImageNet while the final classification layer gets removed and global average pooling produces discriminative feature representations. The system uses an attention-based feature refinement method to improve feature representation followed by a regularized fully connected deep classifier which includes batch normalization and dropout for four-class classification of tumors. Model interpretability is achieved using Grad-CAM++ and attention visualization, enabling clear identification of tumor-related regions and improving model transparency. Experimental evaluation demonstrates high classification performance, achieving accuracies of 98.54% for meningioma, 98.65% for glioma, 99.45% for pituitary tumors, and 99.56% for no-tumor cases. These results indicate that presented TransUNet–DenseNet121 approach offers an accurate, reliable, interpretable method for detection of brain tumors automatically from MRI scanned images.
Related
- ReViT-RF: A hybrid ResNet-50–Vision Transformer with a Radiomic Feature Fusion framework for lung tumor classification and TNM-based stage prediction using CT images
- CERVEX: A Foundation-Model Framework With Built-In Explainable AI for Automated Cervical Cytology Classification From Pap Smear Images
- Explainable vision transformer framework for breast lesion classification with Grad-CAM support
- Hybrid deep learning–driven explainable AI framework for fault detection and classification in smart power grids
- DualXBrNet: A Hybrid Dual-Path Cross-Level Fusion Framework for Multiclass Brain Tumor Classification Using MRI Images
BibTeX
@article{babu2026explainable,
title = {{An explainable hybrid transformer–CNN framework for multi-class brain tumor classification from MRI images}},
author = {Bukkapatnam Rakesh Babu and Vullanki Rajesh and Pardhu Thottempudi and Ravilla Dilli},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-76162-z},
url = {https://www.nature.com/articles/s41598-026-76162-z},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - An explainable hybrid transformer–CNN framework for multi-class brain tumor classification from MRI images
AU - Bukkapatnam Rakesh Babu
AU - Vullanki Rajesh
AU - Pardhu Thottempudi
AU - Ravilla Dilli
PY - 2026
DA - 2026/10/27
JO - Scientific Reports
DO - 10.1038/s41598-026-76162-z
UR - https://www.nature.com/articles/s41598-026-76162-z
AB - The identification of brain tumors through magnetic resonance imaging (MRI) requires effective methods that provide precise results because these results determine clinical choices. The study presents a classification system based on hybrid Transformer-CNN technology which classifies brain tumors into multiple types by using tumor segmentation and deep feature extraction and explainable artificial intelligence. The preprocessing of MRI images involves several steps which include skull stripping and CLAHE-based contrast enhancement and normalization and data augmentation and resizing to create high-quality images that maintain uniformity for model input. The TransUNet architecture performs tumor region segmentation by integrating convolutional neural networks with transformer-based attention systems which enable it to recognize both nearby and distant contextual details. The system generates segmentation masks which enable region-of-interest (ROI) feature extraction through DenseNet121 that has been pretrained on ImageNet while the final classification layer gets removed and global average pooling produces discriminative feature representations. The system uses an attention-based feature refinement method to improve feature representation followed by a regularized fully connected deep classifier which includes batch normalization and dropout for four-class classification of tumors. Model interpretability is achieved using Grad-CAM++ and attention visualization, enabling clear identification of tumor-related regions and improving model transparency. Experimental evaluation demonstrates high classification performance, achieving accuracies of 98.54% for meningioma, 98.65% for glioma, 99.45% for pituitary tumors, and 99.56% for no-tumor cases. These results indicate that presented TransUNet–DenseNet121 approach offers an accurate, reliable, interpretable method for detection of brain tumors automatically from MRI scanned images.
SN - 2045-2322
ER -