Medical Centers Utilize AI and Federated Learning for Enhanced Cancer Detection
A committee of experts from leading U.S. medical centers and research institutes is leveraging NVIDIA-powered federated learning to enhance cancer detection through advanced AI models, according to the NVIDIA Blog.
Federated Learning in Medical Imaging
Federated learning is a technique that develops more accurate and generalizable AI models by training on data from diverse sources while maintaining data security and privacy. This method allows multiple organizations to collaborate on AI model development without sensitive data leaving their servers.
John Garrett, associate professor of radiology at the University of Wisconsin–Madison, highlighted the growing complexity of data sharing due to privacy constraints. He emphasized the necessity of federated learning to build and test models across multiple sites simultaneously. Garrett is a member of the Society for Imaging Informatics and Medicine (SIIM) Machine Learning Tools and Research Subcommittee, which collaborates with NVIDIA on federated learning projects since 2019.
“Federated learning techniques allow enhanced data privacy and security in compliance with privacy regulations like GDPR, HIPAA, and others,” noted committee chair Khaled Younis. “Additionally, we see improved model accuracy and generalization.”
Project Collaboration and Support
The project involves collaborators from Case Western, Georgetown University, the Mayo Clinic, the University of California, San Diego, the University of Florida, and Vanderbilt University. They utilize NVIDIA FLARE (NVFlare), an open-source framework with robust security features and advanced privacy protection techniques.
Through the NVIDIA Academic Grant Program, the committee received four NVIDIA RTX A5000 GPUs, distributed across participating research institutes. Additional collaborators used NVIDIA GPUs in the cloud and on-premises servers, demonstrating NVFlare's flexibility.
Cracking the Code for Federated Learning
Six participating medical centers provided data from approximately 50 medical imaging studies each, focusing on renal cell carcinoma, a type of kidney cancer. Yuankai Huo, assistant professor of computer science at Vanderbilt University, explained that federated learning involves exchanging model parameters rather than data during training.
In this framework, an initial global model broadcasts parameters to client servers, which set up local versions trained on proprietary data. Updated parameters from local models are then aggregated to produce a new global model. This cycle repeats until the model's predictions no longer improve.
The group experimented with model architectures and hyperparameters to optimize training speed, accuracy, and the number of imaging studies required for desired precision.
AI-Assisted Annotation with NVIDIA MONAI
Initially, the training data was labeled manually. For the next phase, the team is using NVIDIA MONAI for AI-assisted annotation to compare model performance with AI-segmented versus traditionally annotated training data.
Garrett noted the challenges of non-uniform data across different sites. By retraining the model with MONAI-assisted annotations, the team aims to improve overall annotation accuracy. MONAI Label, an image-labeling tool, reduces the time and effort needed to create new datasets.
Data for both manual and AI-assisted phases is hosted on Flywheel, a medical imaging data and AI platform integrated with NVIDIA MONAI. The team plans to publish their methodology, annotated datasets, and pretrained model to support future work.
“We’re interested in not just exploring these tools,” Garrett said, “but also publishing our work so others can learn and use these tools throughout the medical field.”