AI technologies incorporated into diagnostic imaging tools have proven useful in eliminating confirmation bias, often outperforming human clinicians who may bring their own prejudices. Another issue slowing progress is the need for machine learning frameworks that can access patient data without violating strict privacy regulations.
Hence, chip vendors such as Nvidia have promoted new AI-powered tools for healthcare in general and medical imaging in particular. Nvidia is expanding its Clara AI platform aimed at radiologists to include data protections that allow clinicians to collaborate while maintaining patient privacy.
The GPU leader unveiled a “federated” learning tool for Clara during this week’s Radiology Society of America conference designed to promote collaboration while protecting patient privacy. Clara federal learning runs on Nvidia’s recently announced EGX edge platform.
The “AI with privacy” feature is billed as a reference application for distributed model training while ensuring patient data remains with healthcare providers. For example, Nvidia said distributed clients can train deep learning models locally via edge servers, then collaborate to develop accurate “global models” that can be used by clinicians.
The EGX platform provisions a federated server and clients, providing federal learning applications and the foundational AI model via containers, Nvidia said. Participating hospitals would then provide their own labeled data that are shared on imaging tools. The platform also cuts the time radiologists spend on data labeling.
Collaborators then us EGX servers to train a global model using local data, with the results shared back to a federated learning server via a secure link. “This approach preserves privacy by only sharing partial model weights and no patient records in order to build a new global model through federated averaging,” Nvidia said in a blog post announcing the radiology tools.
While tools like IBM Watson have so far made few inroads in medical segments like drug discovery, healthcare experts note that machine learning tools are proving useful in clinical settings where confirmation bias can hinder patient treatment and outcomes.
Nvidia is among a growing list of AI specialists targeting medical imaging markets as groups like the American College of Radiology (ACR) and university partners expand networks designed to build, validate and share diagnostic models.
ACR is deploying the Clara federated learning tools in its AI-LAB, a national platform for medical imaging. The goal is to accelerate the adoption of AI technologies in clinical practice, the partners said.
Along with Nvidia, the research network also is working with hardware vendors such as Dell Technologies, Hewlett Packard Enterprise, Lenovo and Supermicro to deploy secure edge servers used to handle imaging data.
Nvidia is also partnering with King’s College London and others to develop a federal learning platform for the U.K.’s National Health Service.