What question did this study set out to answer?

To develop an automated and explainable framework for brain tumour detection and classification using MRI images.

April 16, 2026Open Access

Brain Tumour Detection And Multiclass Classification Using Ensemble Deep Learning And Vision Transformers With Explainable AI

Key Points

To develop an automated and explainable framework for brain tumour detection and classification using MRI images.
Used several pre-trained CNN architectures for feature extraction from MRI images.
Utilized a dataset of 3264 MRI images classified into four tumour categories.
Applied data augmentation techniques to enhance model generalization.
Created an ensemble model integrating VGG16, InceptionV3, and Xception to improve accuracy.
Implemented Vision Transformer models for comparative analysis and used LIME for interpretability.
The ensemble model demonstrated improved classification accuracy over individual models.
Xception, VGG16, and InceptionV3 models were identified as the top performers.
The use of Explainable AI provided insights into tumour-affected regions, aiding medical interpretation.

Abstract

Early detection and accurate classification of brain tumours are critical for effective treatment planning and improved patient survival. Magnetic Resonance Imaging (MRI) is widely used for brain tumour diagnosis; however, manual inspection of MRI scans by medical experts is time-consuming and may produce inconsistent results due to variations in human interpretation. To address these limitations, this study proposes an automated deep learning framework for brain tumour detection and multiclass classification using MRI images. The proposed system leverages transfer learning with several pre-trained Convolutional Neural Network (CNN) architectures, including VGG16, VGG19, ResNet50, InceptionV3, InceptionResNetV2, and Xception, to extract meaningful features from MRI images. A dataset containing 3264 MRI images across four categories—Glioma tumour, Meningioma tumour, Pituitary tumour, and No tumour—is utilized, and data augmentation techniques are applied to increase the dataset size and improve model generalization. Based on experimental performance, the three best-performing models—VGG16, InceptionV3, and Xception—are integrated into an ensemble model named IVX16, which combines predictions to enhance classification accuracy and reduce overfitting. In addition, Vision Transformer (ViT) based models such as SWIN, Compact Convolutional Transformer (CCT), and External Attention Network (EANet) are implemented for comparative analysis. To improve transparency and reliability in medical decision-making, Explainable Artificial Intelligence (XAI) techniques, specifically Local Interpretable Model-Agnostic Explanations (LIME), are applied to highlight the tumour-affected regions in MRI images and validate model predictions. Experimental results demonstrate that the proposed ensemble framework achieves superior performance compared to individual deep learning models. Overall, the proposed approach provides an accurate, reliable, and explainable solution for automated brain tumour detection and classification, which can assist healthcare professionals in faster and more consistent clinical diagnosis. \\

KI fragen

Bookmark

View Full Paper