Related Experiment Video
Updated: May 11, 2025
09:53
Quantifying the Brain Metastatic Tumor Micro-Environment using an Organ-On-A Chip 3D Model, Machine Learning, and Confocal Tomography
Published on: August 16, 2020
6.8K
Explainable deep stacking ensemble model for accurate and transparent brain tumor diagnosis.
Rezaul Haque1, Mahbub Alam Khan2, Hamdadur Rahman3
1Department of Computer Science and Engineering, East West University, A, 2 Jahurul Islam Ave, Dhaka, 1212, Bangladesh.
Computers in Biology and Medicine
|April 18, 2025
Summary
This study introduces a robust AI model for brain tumor detection in MRI scans, achieving high accuracy and providing interpretable results for clinical use.
Area of Science:
- Medical Imaging and Artificial Intelligence
- Oncology and Radiology
Background:
- Early brain tumor detection in MRI is crucial for treatment success.
- Deep learning models struggle with limited data diversity, class imbalance, and interpretability.
- Existing studies often use small, single-source datasets without combining diverse feature extraction.
Purpose of the Study:
- To develop a robust and explainable stacking ensemble model for multiclass brain tumor classification.
- To enhance feature aggregation and classification accuracy by combining multiple deep learning architectures.
- To improve the reliability and clinical applicability of AI in brain tumor diagnostics.
Main Methods:
- A stacking ensemble model integrating EfficientNetB0, MobileNetV2, GoogleNet, and Multi-level CapsuleNet with CatBoost meta-learner.
- Creation of two large MRI datasets by merging data from BraTS, Msoud, Br35H, and SARTAJ.
- Application of Borderline-SMOTE, data augmentation, PCA, and Gray Wolf Optimization (GWO) to address class imbalance and extract features.
Main Results:
- Achieved 97.81% F1 score and 98.75% PR AUC on M1, and 98.32% F1 score with 99.34% PR AUC on M2.
- Demonstrated superior performance compared to state-of-the-art CNNs, Vision Transformers, and other ensemble methods.
- Developed an explainable AI (XAI) web-based tool for clinical interaction and visualization of decision-critical regions.
Conclusions:
- The proposed ensemble model offers a reliable, scalable, and efficient solution for brain tumor classification.
- The integration of diverse deep learning models and advanced optimization techniques enhances diagnostic accuracy and robustness.
- The developed XAI tool bridges the gap between advanced AI and clinical practice, aiding in early and accurate brain tumor detection.
