ACR engages in collaborations for AI development with launch of AI-LAB platform
April 09, 2019
by John R. Fischer
, Staff Reporter
The American College of Radiology has struck up a number of collaborations following the launch of its Data Science Institute ACR AI-LAB platform, a software program that enables radiologists of all levels to develop, validate and use artificial intelligence tools within their own facilities for their local clinical needs.
Among its collaborators are Nvidia, GE and Nuance, all of which plan to integrate their own solutions with the vendor-neutral platform to help accelerate the process for creating algorithms and expand the reach of ACR AI-LAB to a greater number of providers.
“Right now, AI for radiology is being driven by research that is happening primarily at institutions with extensive informatics and data science resources, and primarily using single institution patient data. Most radiologists – who do not have the data science training to master AI programming – are unable to participate,” Bibb Allen Jr., M.D., FACR, ACR DSI chief medical officer, told HCB News. “ACR AI-LAB is designed for radiologists with minimal programming experience to use their domain expertise to develop AI tools to improve the care they provide their patients.”
A free, open-source software platform, ACR AI-LAB will be accessible among more than 38,000 ACR members and other radiology professionals to build, share, locally adapt and validate AI algorithms, while preserving the protection of patient data at local institutions.
A part of its NVIDIA Clara Developer platform, the NVIDIA Clara AI toolkit is one such solution that will be integrated with the AI-LAB, providing libraries for data annotation, model training, adaptation and federation, and large-scale deployment. Its pairing with the software follows a three-month pilot program that helped both ACR and NVIDIA determine the requirements and direction that would enable facilities to work together and with other industry stakeholders to refine AI algorithms without sharing potentially sensitive patient data.
Another integrated technology is GE Healthcare’s Edison AI platform, which is powered by the NVIDIA Clara and allows its users to trace data as they develop an algorithm, an asset which could help ACR members simplify their creation of compliant AI applications. It also enables radiologists to combine multiparametric data sets and clinical information from the EHR to form more accurate algorithms. By integrating with the ACR AI-LAB, the solution will allow hospital executives to add continuous value to their installed medical devices with smart workflows, and developers to combine data from different modalities, vendors and care settings to simplify transitions from AI research to productive AI usage.
Clara is also a power source of Nuance’s AI Marketplace solution, which provides direct access to more than 70 percent of radiologists across 5,800 connected healthcare facilities. In pairing its solution with ACR AI-LAB, Nuance will allow its users to develop and scale up AI solutions for their practices. AI Marketplace users who employ PowerScribe One can also access the largest storefront of AI imaging algorithms and integrate them directly into their existing reporting workflows to form a collective feedback channel.
ACR AI-LAB leverages existing ACR assets including DART (Data archive & research toolkit) and TRIAD (Transfer of Images and Data), which connects thousands of radiological practices for the ACR accreditation and registry programs and enables ACR to deliver AI computational solutions to institutions interested in developing AI algorithms on premise. Such assets are predicted by Allen to play a large role in the success of the platform.
“Data scientists can become engaged by providing "base" algorithms that can be used, trained and modified by radiologists at each site,” he said. “Eventually, we anticipate that multiple institutions and AI developers will begin working together using transfer learning to develop algorithms that can be broadly generalized to many clinical practices. By providing tools to aggregate and annotate cases, run and evaluate AI algorithms and collaborate with developers and other institutions on training AI models, we enable radiology professionals from all backgrounds to become involved.”
Demonstrations of the initial version of ACR AI-LAB will take place from May 18-22 at the 2019 ACR Annual Meeting in Washington, D.C.