Related Experiment Video
Updated: Sep 17, 2025
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
48.4K
Hybrid transfer learning and self-attention framework for robust MRI-based brain tumor classification.
Soumyarashmi Panigrahi1, Dibya Ranjan Das Adhikary2, Binod Kumar Pattanayak1
1Department of Computer Science & Engineering, Siksha 'O' Anusandhan Deemed to be University, Bhubaneswar, India.
Scientific Reports
|July 2, 2025
Summary
This study introduces a hybrid deep learning model for accurate brain tumor classification from MRI scans, achieving 99.41% accuracy. The model enhances diagnostic reliability and transparency, improving patient outcomes.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computational Biology
Background:
- Brain tumors are a leading cause of cancer mortality globally.
- Accurate and timely diagnosis is critical for improving patient survival rates.
- Manual analysis of Magnetic Resonance Imaging (MRI) for brain tumor detection is time-consuming and prone to errors, necessitating automated solutions like Computer-Aided Diagnosis (CAD) systems.
Purpose of the Study:
- To develop and evaluate a novel hybrid deep learning model for enhanced brain tumor classification using MRI images.
- To improve the accuracy, efficiency, and interpretability of automated brain tumor detection systems.
- To leverage Transfer Learning (TL) and attention mechanisms for superior feature extraction and classification performance.
Main Methods:
- A hybrid model, DenseTransformer, was proposed, combining features from the DenseNet201 Convolutional Neural Network (CNN) with a Transformer architecture.
- The model incorporated Multi-Head Self-Attention (MHSA) and Squeeze-and-Excitation Attention (SEA) blocks to refine feature representation.
- Performance was evaluated on the Br35H dataset (3,000 MRI images) and compared against other pre-trained models (VGG19, InceptionV3, Xception, MobileNetV2, ResNet50V2).
- Statistical validation was performed using Z-test, DeLong's test, and McNemar's test.
- Explainable AI (XAI) techniques, including Gradient-weighted Class Activation Mapping (Grad-CAM) and Local Interpretable Model-agnostic Explanations (LIME), were employed for model interpretability.
Main Results:
- The proposed DenseTransformer model achieved a high and consistent accuracy of 99.41% in brain tumor classification.
- Statistical analyses confirmed the model's reliability and superior performance.
- XAI techniques provided transparency into the model's decision-making process, enhancing trust and interpretability.
Conclusions:
- The developed hybrid deep learning model demonstrates significant potential as a reliable and accurate tool for brain tumor diagnosis.
- Integrating Transfer Learning and attention mechanisms effectively addresses challenges in MRI analysis, such as computational intensity and noise sensitivity.
- This advancement in AI-driven medical imaging can contribute to improved patient outcomes and more efficient clinical workflows.
