NVIDIA Brings AI To Health Care While Protecting Patient Data

eWEEK HEALTH-CARE TREND ANALYSIS: At the RSNA conference, NVIDIA demonstrates secure, distributed AI modeling for health-care use cases.

Download the authoritative guide: Big Data: Mining Data for Revenue

NVIDIA.sign

Health care has been one of the early adopters of artificial intelligence (AI), because the technology has the ability to find needles in haystacks of data much faster than people can. This increase in speed can often save lives; time is of the utmost importance in this industry. Also, AI systems can often find things that are not apparent to even the most skilled clinician.

As an example, ZK Research recently interviewed a data scientist at a leading health-care institution in the Boston area where the radiology department used AI to inspect MRIs. AI systems can spot brain bleeds and other issues that are small and imperceptible to the human eye. This enables doctors to spend more time treating patients and less time diagnosing the problem.

Patient data privacy limits the use of AI in health care

One of the biggest factors holding AI in health care back is enabling machine learning and AI frameworks to access the massive volumes of patient data without violating strict privacy violations. At the recent annual Radiological Society of North America (RSNA) conference, NVIDA demonstrated a solution that can get over this hurdle.

NVIDIA introduced its Clara Federated Learning, which uses a distributed, collaborative learning technique that keeps patient data inside the walls of a health-care provider instead of pulling it into a cloud service. This is accomplished by running Clara Federated Learning on the recently announced NVIDIA EGX intelligent edge computing platform.

NVIDIA, the industry’s GPU market leader, has been instrumental in bringing machine learning and AI into more verticals by building systems to address specific industry needs, and the new health-care use case is a great example. Clara Federated Learning (FL) is a reference application for distributed, collaborative AI model training that ensures privacy for patient information. The workload on an edge server from any number of NVIDIA partners can train systems globally by sharing labeled data with other hospitals. The larger data set creates more accurate models and significantly reduces the time clinicians need to spend labeling data.

Clara Federated Learning speeds up AI training while protecting patient data

The system has been packaged into a Helm chart to make it easier to deploy on Kubernetes Infrastructure. The NVIDIA EGX Edge compute node securely provisions the federated server and the collaborating clients, delivering the full stack of what is needed to run a federated learning project, including containers and the initial AI model.

The uniqueness of the systems is that it uses distributed training data across multiple health-care institutions to developer better AI models without sharing patient data. Each hospital can label its own patient data using the NVIDIA Clara AI-Assisted Annotation SDK, which has been integrated into a number of medical viewers such as 3D slicer, MITK, Fovia and Philips Intellispace Discovery. The pre-trained models and transfer learning techniques dramatically speeds up the learning time. Some hospitals have told ZK Research that processes that used to take hours can now be done in minutes, providing a huge boost to the organization.

The privacy is ensured as the training results are shared back to the federated learning server over a secure link. Also, the system only shares model information and not patient records, protecting sensitive information. The process runs iteratively until the AI Model reaches a predetermined level of accuracy. The distributed model accelerates learning through the use of a larger data set but keeps patient information secure and private.

United States, United Kingdom are leading the charge

The system is being developed in conjunction with a number of leading health-care organizations in the U.S. and UK. This includes the American College of Radiology, Massachusetts General Hospital, Brigham and Women’s Hospital Center for Clinical Data Science and UCLA Health, with the goal of developing personalized AI for their doctors and patients.

In the UK, NVIDIA has partnered with King’s College London and Owkin to create a federated learning platform for the National Health Service. The Owkin Connect platform running on Clara FL enables algorithms to be used in multiple hospitals. In this case, blockchain is being used a distributed ledger to capture and trace data used for modeling.

AI will change the world in ways never imagined, and there is no better use case than in health care. Massive amounts of data are created by the health-care system today, but there is no mechanism to connect the dots to find key insights. NVIDIA’s Clara FL allows for those dots to be connected--and across historically disparate islands without compromising patient privacy.

Zeus Kerravala is an eWEEK regular contributor and the founder and principal analyst with ZK Research. He spent 10 years at Yankee Group and prior to that held a number of corporate IT positions.