US researchers are using federated learning, powered by NVIDIA technologies, to improve AI models in tumor segmentation while ensuring data privacy.
Several of the main medical and research centers in the U.S. have started to use NVIDIA and federated learning to develop artificial intelligence models (AI) advanced, focused on the tumor segmentation.
This technique allows multiple institutions to collaborate on improving AI models without compromising data security or privacy, an increasingly urgent need in the healthcare field.
Federated learning: a key tool for medical AI
He federated learning It is an innovative technique that allows AI models to be trained without the data needing to leave the local servers of the participating institutions.
This approach solves many of the challenges related to privacy and regulatory compliance, such as General Data Protection Regulation (GDPR) and the U.S. Health Insurance Portability and Accountability Act (HIPAA), which regulate the use of personal and medical data.
According to John Garrett, associate professor of radiology at the University of Wisconsin-Madison, the technique responds to the growing complexity in medical data management: “Adopting federated learning to build and test models across multiple sites at once is the only practical way to keep up.”
This project brings together institutions such as Case Western Reserve University, Georgetown Universitythe Mayo Clinicthe University of California, San Diegoamong others, in an effort to advance the application of medical AI.
NVIDIA FLARE’s role in collaboration
To facilitate this process, the team uses NVIDIA FLARE (NVFlare), an open source platform that offers advanced security and privacy features.
Through the NVIDIA Academic Fellowship Program, researchers received GPUs NVIDIA RTX A5000which were distributed among the different participating centers to establish workstations dedicated to federated learning.
This approach also highlights the flexibility of NVFlare, allowing both local and cloud GPUs to be used for training.
AI-assisted optimization: NVIDIA MONAI
In the first phase of the project, the data used to train the AI models was manually annotated. However, in the next stage, the team will employ NVIDIA MONAI to implement AI assisted annotation.
This method allows models to be optimized by automatically segmenting medical images, which is expected to speed up the process and offer a comparison with traditional annotation methods.
“The biggest challenge in federated learning activities is often that the data at each site is not uniform,” says Garrett. “Each location uses different equipment and protocols, which makes consistency difficult. With MONAI, we will evaluate whether this tool improves annotation accuracy in a federated environment.”
The team is using MONEY Labela tool that makes it easy to create custom applications for image annotation, significantly reducing the time and effort required to generate new datasets.
Experts in the field will validate the AI-generated segmentations before using them to train federated learning models.
Methodology and future results
Once the project is complete, the researchers plan to publish their methodology, annotated data sets, and pre-trained models so that they can be used by other teams and so that the medical community can take advantage of these advances in AI development.
This collaborative approach has the potential to provide new tools and improve processes in the tumor segmentation by using more accurate and generalizable AI models, while complying with privacy regulations and protecting the security of sensitive data.
Source: geeksroom.com
