Objective Assessment of Endoscopic Tip Control Using Artificial Intelligence and Video Motion Analysis: A Proof-of-Concept Study

Background and Aims Precise control of the endoscope's distal tip is a core visuomotor competence underpinning procedural safety and success. Despite its importance, objective assessment of tip control remains limited by reliance on subjective human judgment. We aimed to develop and internally validate a computer-vision pipeline, combining an artificial intelligence (AI)-based classifier of on- and off-target task engagement with optical-flow motion analysis, for objective quantification of endoscopic tip control within a standardized training model. Methods Thirteen endoscopists of varying experience performed standardized snare-tip soft coagulation tasks on stencil-imprinted ham models. Procedures were analysed using a computer-vision pipeline combining AI-based on-target/off-target classification with optical-flow motion analysis. Blinded raters assigned subjective tip control scores (0–100%). Associations between pipeline-derived motion metrics and scores were examined using generalized linear mixed models, followed by model selection and internal cross-validation. Results Inter-rater reliability of subjective tip control scores was moderate-to-good (ICC[A, k̂=3]: 0.74 (95% CI 0.60-0.87)). Multivariable analyses consistently identified three key predictors of tip control. Higher off-target time was associated with a 10.1 percentage-point reduction in tip control scores (95%CI −13.7 to −6.5), greater motion volatility with a 14.7-point reduction (95%CI −22.2 to −7.3), whereas larger volatility peaks were associated with a 12.3-point increase in scores (95%CI 4.2 to 20.5). The final model demonstrated good out-of-sample performance (mean absolute error: 12.2, root mean squared error: 15.0). Conclusions In this single-centre proof-of-concept study, AI-based labelling of task engagement combined with optical-flow motion analysis approximated panel rated tip control, which was characterized by sustained spatial accuracy, global motion stability, and well-timed corrective adjustments. This concept could support the development of a scalable and standardized tool for evaluating endoscopic tip control.