Latest / Medsolve Dynamics / Why AI Is Not a One-Time Purchase in Radiation Oncology
Transcript
- Yoel Bakas: Your physicist knows that your LINAC drifts, right? That's the job. Daily QA, weekly QA, monthly checks. And all of these are done against hard tolerance tables. That's because a LINAC wanders in mostly known, expected ways. And your physicist is the person that you pay to make sure those deviations are caught. Now you buy an AI tool. It turns out it also drifts. But it drifts differently from expected parameters. And that drift is the whole problem. There is no TG-142 to model against it. When the vendor updates the model, your population shifts, your imaging protocol changes, the model that you validated in January is now different. and no one is checking it against the tolerance table because one doesn't exist. The drift is unbounded and it's silent. So that AI tool you bought is actually a recurring monitoring burden. Someone qualified has to continue to confirm the tool is doing what it's intended to do. And they have to do this for as long as you run it. Sometimes that's physicist time, but at the end of the day, it's an operational question. Who is funded for and actually maintaining this responsibility inside your program And ensuring the tool you originally bought is working the same expected way at month nine. That ongoing validation is part of the work that we do at Medsolve. And of course, we do it before you ever sign the AI deal. Not whether it worked at the demo. whether your program funded the work to make sure it's still working at month nine. Before you buy the AI, take into account the cost of ongoing validation. Not just the price on the quote.