NVIDIA announces AI partnerships with OSU, NIH
November 30, 2018
by Lisa Chamoff
, Contributing Reporter
In its second year exhibiting at RSNA, computing company NVIDIA highlighted its collaborations with healthcare providers, OEMs, startups and even the federal government on applying artificial intelligence (AI) to medical imaging.
This includes a new partnership with Ohio State University that allows it to take advantage of NVIDIA’s new Clara software development kit to build an in-house AI marketplace for clinical imaging.
“They’re a really great partner because they’ve been thinking about this and they’ve been actually building their own artificial intelligence, and now they’re at the point of wanting to deploy it,” Kimberly Powell, vice president of health care at NVIDIA, told HCB News during an interview at the show.
At NVIDIA’s booth in the Machine Learning Showcase section of the show floor at McCormick Place, OSU’s Laboratory for Augmented Intelligence in Imaging showcased an algorithm that prioritizes critical images that need to be read, such as a femur fracture and a coronary artery calcification.
Matthew Bigelow, a biomedical imaging consultant at OSU, recounted a situation in which a radiologist thought an image looked “textbook normal,” was interrupted by a phone call, and then fortunately decided to give the image another read and found an abnormality.
“If he had the ability to just click a button and have it highlight immediately somewhere he may have missed, then why not have that extra set of eyes?” Bigelow said.
Bigelow explained that NVIDIA’s Clara platform and inference server provides the infrastructure to allow OSU to deploy the algorithms they build internally. He compared the inference server on NVIDIA’s Clara platform to a computer and the models to pieces of software. Not having the server would be akin to needing to buy separate computers for reading emails, web browsing and word processing.
“It gives us one place to place all of these models,” Bigelow said. “You don’t have to build a new machine to do one specific thing. This allows us, as we create a model, or find a model we want to use, to just insert it into one machine and utilize all of our GPU resources as efficiently as possible.”
Powell said OSU is one of many institutions that will be creating their own tailored AI.
“Every radiology practice is different at different institutions,” Powell said. “We believe that the future is going to be, people are going to want to build their own custom AI. There’s going to be, probably, thousands of algorithms, because it’s everything from the order entry to the setting up of the machines to placing the patients, to the quality assessment of the image that was captured, to all of the downstream processing. The opportunities are countless right now and we’re still at the early stages.”
NVIDIA also announced this week that it is partnering with the National Institutes of Health to bring AI tools to clinical trials.
This would include using AI to look at how patients are responding to certain drugs using advanced imaging. Powell said they plan to see research coming out in the next six months.
“Being able to measure that response in quantitative ways is pretty arduous,” Powell said. “So, we’re really thinking about how can AI play a better role in better staging of patients and a better understanding of their response. I think it’s going to be an area of radiology where it’s so impactful to what we do in research, as well as what the drug companies need to get better at. And I think patients will benefit.”