Video-based analysis of camera–instrument distance dynamics and operative efficiency in expert and novice robot-assisted partial nephrectomy

In robot-assisted partial nephrectomy (RAPN), experts and novices differ in many metrics, but it is unknown which difference relates to novice operative efficiency and whether it improves with early experience. We addressed these questions using routinely archived two-dimensional videos. We analyzed 155 RAPN by one expert (65) and eight novices (90). Deep learning tracked the right-hand instrument across 300,000 frames. In four operative phases (bowel mobilization to tumor resection), we computed working-tip kinematics (speed, path length, direction reversals), the tip's position in the surgeon's view, and the camera-instrument distance, reconstructed using the 8-mm shaft as a size reference. Distance stability was expressed as the standard deviation (SD) of the four phase medians (phase-to-phase distance SD). Console-time associations in novices were assessed by Spearman correlation and logistic regression adjusted for R.E.N.A.L. score and body mass index. The expert moved the working tip faster, with fewer direction reversals, more consistent tip placement in the view, and a more stable camera-instrument distance across phases (phase-to-phase SD 5.5 vs. 7.8 mm, P < 0.001). Within novices, the phase-to-phase SD was the only video-derived metric associated with console time (ρ = 0.41; adjusted odds ratio 1.24 per mm, 95% confidence interval 1.04-1.48). With experience, reversals decreased and median distance increased, but the phase-to-phase SD showed no detectable change. Among expert-novice differences, only the phase-to-phase stability of the camera-instrument distance was associated with console time in novices and showed no detectable change during early experience. Distance stability is a candidate training target that warrants prospective validation.