Marc Kohli, MD, Medical Director of Imaging Informatics, UCSF Health. Hosted by Mike Moore. Runtime 29:22.
Ask most health systems what they are doing about AI in imaging and the answer is a list of models. Marc Kohli starts somewhere less exciting and, he argues, more decisive: the routing and the standards underneath.
Kohli is Medical Director of Imaging Informatics at UCSF Health. He and Mike were on a panel together at SIIM in Pittsburgh this summer on getting imaging past the monolith, and this conversation picks up where that left off. UCSF has been deliberately conservative about AI, starting from a clinical problem rather than from a wish to use the technology. The system doing the most for their AI program turns out to be a DICOM router installed almost ten years ago, an insight Kohli credits to his colleague John Mongan, which now works as their AI orchestrator.
The sharper part of the conversation is about who gets left out. Imaging training data is not representative of patients across the country, and Kohli is specific about the consequence: critical access hospitals and rural populations are the most likely to be missed. A model works best on patients who resemble the ones it learned from, and nothing about that gap announces itself in the reading room.
15:04 The AI orchestrator was already in the building. "We put in a DICOM router almost ten years ago... Being able to do that in a rules-based way is incredibly valuable. We basically use it as our AI orchestrator." Bought long before anyone was buying AI, and now the thing that makes AI practical. Kohli credits the insight to his colleague John Mongan.
7:17 "Communication is a potentially lossy compression process." The best one-line case for Common Data Elements we have heard. A concept goes into your head, words come out, and the other person has to decompress them and hope they land in the same place.
20:24 Who gets missed. "The other population at high risk of losing out is the people in critical access hospitals and rural populations... you're still going to have things benefiting people on the coasts and kind of missing the middle of the country."
Marc Kohli, MD, is Medical Director of Imaging Informatics at UCSF Health. He is a former chair of the SIIM board, co-chairs the RSNA and ACR Common Data Elements effort, and contributed to RSNA's "AI for Radiology: A Primer." He still reads clinically every week alongside his informatics work.
The plumbing decides who your imaging AI works for covers the routing and standards argument in detail, plus three questions for working out whether your own AI plan has a plumbing problem underneath it.
Flight Path is Mach7's show about where healthcare imaging is going and the people helping it get there, hosted by Mike Moore. New episodes every other Tuesday.
Follow so you don't miss the next one: YouTube · Spotify · Apple Podcasts