It is impossible for me to sit and contemplate artificial intelligence (AI) without a voyage into my psyche and scenes from two movies. Scene 1, Terminator: Skynet run amok, trampling the humans underfoot. Scene 2, 2001: A Space Odyssey: HAL looking at me with that single, unblinking red eye. “I don’t think I can do that, Doug.”
I have colleagues who are deep into the AI radiology angle. Regardless of who you read, who you listen to, or who you watch, there is a widespread opinion that radiology is going to be early in the takeover phase of the AI era. Our digital information and reliance on images that depict either normal or abnormal seems to readily allow input into a smart computer that analyzes our data and does what a human does: say, “Can’t exclude tumor,” and send a bill.
It correlates what it reviews with a huge repository of normals and abnormals, finds one that looks the same, and misdiagnoses the normal variant as a fracture.
Look, I’m concerned about this, too. I have participated in a few of these AI studies and two things worry me. One, the algorithms can be pretty good. The machine can call a lot of basic pathology quickly and accurately. Just not always. And two, the learning curve for these systems is pretty fast. They aren’t getting worse, they are getting better. Rapidly.
So, perhaps it is best to think about the things we offer (our “value-added” for you MBAs and venture capital types) instead of how nice it will be when the friendly radiologist is replaced by a server in the back of the IT room. Here’s a short list:
• Who’s going to give the ordering doctor chocolate or gummy bears when they come into the reading room for a consult, huh?
• Who’s going to take the call from the technologist about the patient who got vasovagal and hit their elbow on the door frame after their MRI scan?
• Who’s going to explain the vagaries of that report the machine issued with the patient who wants to talk to the reading physician (or computer)?
• Who’s going to speak live with the surgeon in the OR when they can’t find the lesion the computer called on the study?
On the one hand, I think we will likely survive and flourish through this whole thing. But on the other, I will tell you this: it is nice to be eyeing retirement and likely not having to deal with HAL refusing to let me in the building or a T800 with a plasma rifle in the hallway.
Keep doing that good work (and don’t mind that new computer in the server room). I’ll be back. Mahalo.Back To Top
Phillips CD. Wet Read: AI and Me. Appl Radiol. 2021;50(5):64.
Dr. Phillips is a Professor of Radiology, Director of Head and Neck Imaging, at Weill Cornell Medical College, NewYork-Presbyterian Hospital, New York, NY. He is a member of the Applied Radiology Editorial Advisory Board.