Explainable AI in medical imaging: an interpretable and collaborative federated learning model for brain tumor
Qurat-Ul-Ain Mastoi1, Shahid Latif1, Sarfraz Brohi1
1School of Computing and Creative Technologies, University of the West of England Bristol, Bristol, United Kingdom.
Frontiers in Oncology
|March 14, 2025
Summary
This study introduces a collaborative federated learning model with explainable AI for accurate brain tumor classification. The model achieves 94% accuracy, offering enhanced transparency and reliability for clinical applications.
Area of Science:
- Artificial Intelligence
- Medical Imaging
- Machine Learning
Background:
- Accurate brain tumor classification is crucial for effective treatment, yet faces challenges with data diversity and model transparency.
- Magnetic Resonance Imaging (MRI) is vital for tumor diagnosis, but deep learning models often struggle with centralized data limitations.
- Existing deep learning models for brain tumor classification lack transparency, hindering clinical trust and adoption.
Purpose of the Study:
- To propose a novel collaborative federated learning model (CFLM) integrated with explainable artificial intelligence (XAI).
- To address limitations in dataset availability and model interpretability in brain tumor classification.
- To enhance the reliability and transparency of AI-driven diagnostic tools for neuro-oncology.
Main Methods:
- Integration of the GoogLeNet architecture with a federated learning (FL) framework for privacy-preserving collaborative training.
- Implementation of Gradient-weighted Class Activation Mapping (Grad-CAM) and saliency map visualizations for model interpretability.
- Utilizing a decentralized approach with 10 clients and 50 communication rounds for training on local datasets.
Main Results:
- The proposed CFLM achieved a classification accuracy of 94% for identifying glioma, meningioma, no tumor, and pituitary tumors.
- Incorporation of Grad-CAM and saliency maps provided meaningful graphical interpretations for healthcare specialists.
- Demonstrated the effectiveness of federated learning in maintaining data privacy while enabling collaborative model training.
Conclusions:
- The developed model offers an efficient and interpretable solution for brain tumor classification.
- The integration of FL and XAI presents a significant advancement for transparent and reliable clinical decision-making.
- This framework holds substantial potential to improve the accuracy and trustworthiness of AI in neuro-oncology.
