Related Experiment Video
Updated: Jun 12, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Segmentation Synergy with a Dual U-Net and Federated Learning with CNNRF Models for Enhanced Brain Tumor Analysis
Vinay Kukreja1, Ayush Dogra1, Satvik Vats2
1Chitkara University Institute of Engineering and Technology, Chitkara University, Punjab, India.
Current Medical Imaging
|September 19, 2024
Summary
This study introduces a federated learning approach combining convolutional neural networks (CNNs) and random forests (RF) with dual U-Net segmentation for accurate brain tumor identification from MRI scans, ensuring data privacy.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Healthcare
- Machine Learning for Diagnostics
Background:
- Brain tumor diagnosis presents significant imaging challenges, requiring precise differentiation between normal and pathological tissues.
- Machine learning techniques offer potential for enhancing brain tumor identification accuracy from MRI data.
- Federated learning addresses data privacy concerns in collaborative model training.
Purpose of the Study:
- To evaluate the efficiency of a federated learning method integrating convolutional neural networks (CNNs) and random forests (RF) with dual U-Net segmentation.
- To assess the performance of this hybrid model for brain tumor identification in preprocessed MRI scans.
- To ensure a privacy-preserving approach for collaborative model training.
Main Methods:
- Utilized federated learning to train a CNN-RF model across diverse datasets while preserving data privacy.
- Applied Median, Gaussian, and Wiener filters for MRI image preprocessing to reduce noise and facilitate feature extraction.
- Employed a dual U-Net architecture for segmentation and evaluated performance using precision, recall, F1-score, and accuracy.
Main Results:
- The hybrid model achieved high classification performance on local datasets, with CRPs ranging from 91.28% to 95.52%.
- Federated averaging resulted in a collective model with 97% accuracy, outperforming individual client models.
- The federated averaging method effectively consolidated individual model insights into a global model while maintaining data privacy.
Conclusions:
- The integrated federated learning framework, CNN-RF hybrid model, and dual U-Net segmentation offer a robust, privacy-preserving solution for brain tumor MRI identification.
- The study demonstrates the technique's promise in improving brain tumor classification quality.
- This approach provides a viable pathway for practical clinical application in neuro-oncology.
