Background?An AI feedback system combined with a bronchial tree phantom enhances novices' performance in identifying bronchial segments. However, simultaneous training and evaluation with a fixed model lacks objectivity. To address this, a novel AI-based feedback system has been developed using tracheal tree models from diverse patients' chest CT scans. This study aims to examine whether this system improves novices' bronchoscopy performance.

Methods?The RCT included bronchoscopy novices (n=30). The control group received written instructions; high-fidelity simulator group used a standard tracheal model; personalized data group trained with CT-generated models. Participants completed two tests: one on the trained standard tracheal tree model and another on a non-training CT-generated model.

Results?The high-fidelity simulator and personalized data groups showed similar performance on the standard tracheal tree model test. However, the personalized data group outperformed the high-fidelity simulator group on the non-training CT-generated model test. They achieved higher diagnostic completeness (P=0.039), greater structured progress (P<0.001), and shorter procedure time (P<0.001).

Conclusions?Combined AI feedback system with personalized data-driven models enhances novices' bronchoscopy performance comprehensively, systematically, and efficiently.