“AI in diagnostics is where the future is going to be. And I think our kids will look back on healthcare now and think, ‘Oh my gosh, radiologists used to just read those images on their own?’” declared Teri Thomas.
Thomas — CEO of Volpara Health, a company that sells software for breast cancer screenings — made this comment during an interview this week at RSNA the Radiological Society of North America’s annual conference in Chicago.
She noted that Volpara’s belief that AI will shape the future of radiology played a major role in the company’s decision to sell itself to Lunit earlier this year. Lunit is a South Korean company specializing in AI tools for cancer diagnostics and therapeutics.
The deal has allowed Volpara to share its extensive knowledge of the U.S. healthcare system with Lunit, and it has helped Volpara strengthen its AI expertise and capabilities, Thomas explained.
Working alongside Lunit leaders has quickly deepened Thomas’ understanding of AI, she said. She highlighted a recent trip she took to Stockholm to visit Saint Göran Hospital, which adopted Lunit’s AI solution for mammography this year.
“The standard for reading mammograms in Europe — and most of the rest of the world outside of the U.S., by the way — is that you have one radiologist who looks at the image and determines whether they think there’s cancer or not, and then they have a completely separate one look at it, unbiased. Then they compare their results, and if they disagree, a third looks at it,” Thomas explained.
Given the ongoing shortage of radiologists, there is a huge opportunity for AI to step in for a process like this, she pointed out. And that’s what happens at Saint Göran.
“They have one radiologist, and then they have the AI essentially be the second radiologist. They compare what the AI found with what the radiologist found, and if there’s a disagreement, then they bring another radiologist in,” she said.
At Saint Göran, the combination of one radiologist plus AI outperforms two radiologists, Thomas declared.
Individual radiologists train by looking at thousands of images, but not millions, she noted. She also pointed out that the way humans analyze images is “less systematic” than the way AI models have been trained.
“It’s like having a different angle. They call it a ‘second read’, but it’s actually a second read that was trained differently and can find different things,” Thomas explained.
The deal has also helped Volpara integrate AI models into its software, she stated.
The company is adding AI tools to help clinicians with various aspects of mammography, such as breast positioning, dosage and compression.
“Some people were taught [to compress] until [the patient] says, ‘Ow stop!’ That’s horrible — there is such a thing as overcompressing, and you actually don’t get as good of an image. There’s commonly also undercompressing, and you also don’t get as good of an image. So applying AI can help figure out what is the optimal compression, the best positioning, that the X-ray machinery is proper, and the dosage of radiation is correct,” Thomas remarked.
By adding these AI functionalities, Lunit and Volpara seek to make it easier and faster for clinicians to feel like they obtained the fullest picture possible, she noted.
Photo: ipopba, Getty Images
