Precision and Robust Models on Healthcare Institution Federated Learning for Predicting HCC on Portal Venous CT
IEEE Journal of Biomedical and Health Informatics
|May 13, 2024
Summary
This study introduces a Hybrid-ResUNet model using federated learning (FL) for accurate liver cancer (HCC) segmentation in CT scans. The approach enhances diagnostic precision while protecting patient privacy in medical AI.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Healthcare
- Oncology
Background:
- Hepatocellular carcinoma (HCC) diagnosis relies heavily on medical imaging interpretation, often requiring specialized expertise.
- Current methods for liver and tumor segmentation in CT scans present challenges in accuracy and efficiency.
Purpose of the Study:
- To develop an advanced deep learning framework for precise segmentation of liver and tumor regions in medical images.
- To leverage federated learning (FL) for privacy-preserving, large-scale medical image analysis.
Main Methods:
- Integration of 2D and 3D deep learning models within a federated learning framework.
- Utilized 131 CT scans from the Liver Tumor Segmentation (LiTS) challenge.
- Proposed a novel Hybrid-ResUNet model for segmentation.
Main Results:
- The Hybrid-ResUNet model achieved a high Dice score of 0.9433 and an AUC of 0.9965.
- Demonstrated superior performance compared to ResNet and EfficientNet models.
- Showcased resilience to data imbalances in the FL context.
Conclusions:
- The proposed FL approach enables accurate HCC segmentation while safeguarding patient data privacy.
- Highlights the potential of collaborative AI models for clinical trials and medical AI development.
- Recommends human-AI collaboration for enhanced feature extraction and knowledge transfer in medical AI.
