Real-time artificial intelligence-based needle tracking for ultrasound-guided regional anesthesia training: a pilot prospective randomized controlled trial

Abstract Background Proper needle visualization is a major technical challenge for novices learning ultrasound-guided regional anesthesia (UGRA). We developed a ‘You Only Look Once version 5’ (YOLOv5)-based system that records quantitative needle trajectory data and delivers real-time visual and auditory feedback to support training. This pilot study aimed to explore whether short-term use of the real-time feedback system improved puncture safety and whether needle oscillation amplitude, a system-derived trajectory-based metric, could serve as an objective indicator of technical proficiency in UGRA training. Methods Twenty-three medical students with no UGRA experience were randomized into control ( n = 12) and YOLOv5 feedback ( n = 11) groups. Participants performed three in-plane punctures of increasing difficulty on phantoms: sagittal (puncture A), coronal (puncture B), and transverse (puncture C). Each included watching an instructional video, a 5-minute practice (with feedback for the YOLOv5 group or without for the control group), and a test puncture without feedback. The outcomes were needle tip disappearance frequency compared between the control and YOLOv5 feedback groups and needle oscillation amplitude — the maximum lateral deviation from the ideal insertion line, automatically recorded by the system — compared between punctures classified as safe and unsafe by an independent blinded assessor. Results Needle tip disappearance frequency did not differ between the control and YOLOv5 feedback groups across all approaches (puncture A: median, 0 vs. 0, P = 0.933; puncture B: 0 vs. 1, P = 0.246; puncture C: 1 vs. 2, P = 0.463). Only puncture C yielded both safe and unsafe punctures, enabling comparative analysis of the needle oscillation amplitude. Participants performing safe punctures (Level 1, n = 9) exhibited significantly smaller needle oscillation amplitudes (136.0 ± 32.1 pixels) than those performing unsafe punctures (Level 3, n = 12, 230.5 ± 85.4 pixels; P = 0.003). Conclusions This pilot study identified needle oscillation amplitude as a candidate objective proficiency metric in UGRA training, with preliminary evidence differentiating safe from unsafe punctures. Short-term training with the current prototype did not produce measurable improvements in puncture safety. However, the usability challenges encountered with this feedback system provided foundational insights for developing an optimized trajectory-based feedback system. Trial registration UMIN Clinical Trials Registry (UMIN000055602), registered October 1, 2024.