Developing effective AI tools for healthcare means working multiple stakeholders and understanding the clinical, according to panelists at GTC in San Jose.
Artificial intelligence has been making its way into nearly every facet of modern life from self-driving cars to tagging photos on facebook. In such a human-driven industry like healthcare, many are skeptical about where AI will fit in.
However, it doesn’t have to be human versus machine— rather, the future can be humans working with machines, according to panelists at the Women in Health AI event at the GPU Technology Conference in San Jose this morning.
In order for the AI tools to work well in the clinical space it’s important to incorporate a diverse group of voices developing those systems.
Want to publish your own articles on DistilINFO Publications?
Send us an email, we will get in touch with you.
“In healthcare, just building an app is not good enough. In fact, it’s useless if you don’t involve people from many, many different backgrounds,” Kimberly Powell, vice president of healthcare at Nvidia, said during the panel. “Just building a neural network that can outperform some benchmark is 100 percent useless in healthcare if it doesn’t come from the right space.”
Powell said it’s key to understand the space that the technology is being developed for. Oftentimes that means starting with a problem and talking to clinicians.
Radiology is one area of medicine where AI has been making headway.
“It starts from having a clinical need. The radiologist identified this need a couple of years ago and it has been a topic in our major meetings where they are trying to educate all of the radiologists about AI,” Dr. Elizabeth Jones, of the National Institutes of Health, said on the panel. “What we find is that the radiologist in their workflow is very frustrated. We take some of the smartest people on the planet, and they are frustrated with their work, they say it doesn’t work well for them. They find they don’t have information at their fingertips. That’s where a light bulb went off — now is the time we have large data sets, AI, deep learning in our own department where we have been investigating AI and deep learning for years. So it all just came together kind of by accident.”
It isn’t about cutting the human out of the equation, so much as giving the human tools to improve care. Sahar Arshad, cofounder and chief operating officer at healthcare AI company CloudMedx, discussed how her tool was most effective when the human worked with AI. She referenced a study her company conducted where physicians took a modified version of the United States Medical Licensing Examination on their own, AI took the exam, and then humans and AI worked together.
“The average score of the humans was 75 percent, whereas the AI was 85 percent. So it out performed the humans,” she said. “When AI and [humans] were working together, the score was 90 percent. So when you work in an augmented manor it can really drive the real outcomes.”
Implementing AI is really about looking at a problem and then finding the tools that could help tackle this issue, according to Karley Yoder, director of product management for AI at GE Healthcare. She compared it to other industries where the technology is already being implemented, as a possible avenue for how AI could work in healthcare in the future.
“I think about the driverless car. None of us are excited about it because it has AI. We are excited about it because my two-hour commute this morning could have been 90 minutes and I could have been productive and there [would have been] no accidents on the road,” Yoder said. “So the skillset that is needed, especially from a product side of AI, is to really understand the problem that needs to be solved and then work with smart data scientists and engineers and then construct how you can use that technology to go and solve that problem.”
Editor’s note: This story is part of our coverage of GPU Technology Conference. Nvidia, which organized the conference, invited MobiHealthNews to the event and paid for travel and accommodations. As always, MobiHealthNews maintains its editorial independence and made no promises to Nvidia, including about the content or quantity of coverage.
Date: March 26, 2019
Source: MobiHealthNews
