New AI Grades Kidney-Stone Surgeons by Watching Their Videos
Could tell trainees when they're truly ready — and mean fewer repeat surgeries.
Kidney stone surgery works like this: a doctor threads a thin tube with a camera up through the urinary tract to reach the kidney, finds the stone, and breaks it up. If the surgeon misses part of the anatomy — leaving stone fragments behind — the patient often needs a second procedure. Experienced surgeons have lower repeat-surgery rates, but until now there has been no objective way to measure how well someone navigates the scope. Judgment calls and gut feelings have done the job.
A research team has built a tool called RAUL that changes that. It takes ordinary video from the surgical camera and reconstructs the exact three-dimensional path the scope took, frame by frame. The clever part: for each practice model, the team first records one slow, careful "reference" run, then measures every later attempt against it. No magnetic tracking sensors, no extra equipment clipped to the instruments — just the video feed that already exists.
On nine practice models (lifelike stand-ins, not real patients), the AI located the scope to within about 0.5 millimeters and successfully tracked 86% of video frames, versus roughly 50% with the standard approach. That extra coverage matters, because a path with big gaps is not much use for judging skill. When the researchers turned those paths into navigation scores, the numbers clearly separated veteran trainees from beginners.
Why it matters beyond the lab: hospitals train surgeons for years, largely on subjective feedback. If a video-only system can grade navigation, training programs could give residents concrete, data-backed feedback, and certification bodies could one day require proof of skill rather than time served. It could also be cheap to adopt, since it runs on footage hospitals already record. The honest caveat: so far it has only been tested on practice models, not on living patients, and real anatomy moves, bleeds, and bends in ways plastic does not. The next step is proving it holds up in the operating room.
- The AI rebuilds the exact path a surgical camera took inside the body using only video — no expensive tracking sensors attached to instruments.
- It tracked 86% of video frames accurately, up from about 50% with older 3D-from-video methods, and was accurate to roughly half a millimeter.
- It clearly told apart experienced and inexperienced trainees on navigation scores, pointing toward objective, video-based skill certification.
Why It Matters
Could make surgeon training fairer and more measurable, potentially meaning fewer repeat kidney stone surgeries for patients.