EdTech & Innovation
AI-Assisted Ultrasound Accelerates Learning Curves in Popliteal Sciatic Nerve Blocks
Wang et alPeerJ

In a pilot randomized educational study conducted between June and August 2025 at Shanghai General Hospital, researchers evaluated whether artificial intelligence (AI)-assisted ultrasound improves the learning curve for novice anesthesiology residents performing popliteal fossa sciatic nerve blocks. Twenty-one inexperienced resident trainees were randomized to receive either AI-assisted ultrasound training (n = 10), which included an ultrasound module alongside ethics and principles instruction, or conventional training (n = 11). assessments focused on image acquisition, image interpretation, debriefing metrics, and trainee feedback.
Cumulative sum (CUSUM) analysis revealed that the AI-assisted group reached an estimated learning curve turning point at approximately 12 attempts, whereas the conventional group failed to reach a comparable turning point within the 15-attempt training protocol. However, no statistically significant between-group differences were detected in participant-level phase-specific success rates across the study (all P > 0.05).
Educational efficiency and engagement metrics favored the AI-assisted cohort. Mean debriefing time was shorter in the AI arm (7.33 ± 0.34 vs 10.65 ± 0.61 minutes, P < 0.0001), and trainees in this group accounted for a significantly greater share of speaking time during post-procedure debriefings (median 69.5% [interquartile range 65.70, 70.55] vs 49.0% [46.70, 49.70], P < 0.0001). Trainees using AI also rated training content (4.7 ± 0.5 vs 3.6 ± 1.0, P = 0.0076), overall satisfaction (4.9 ± 0.3 vs 3.5 ± 0.8, P < 0.0001), and procedural confidence (5.0 ± 0.0 vs 4.5 ± 0.7, P = 0.0216) higher than those in standard training, suggesting potential educational utility for AI guidance in regional anesthesia training.
Key Takeaways
- Group-level CUSUM analysis identified a learning curve turning point at approximately 12 attempts in the AI-assisted cohort, whereas no turning point was reached within the 15-attempt period in the conventional group, despite no statistically significant between-group differences in participant-level phase-specific success rates (all P > 0.05).
- AI-assisted training resulted in shorter participant-level mean debriefing times (7.33 ± 0.34 vs 10.65 ± 0.61 minutes, P < 0.0001) and a higher proportion of trainee speaking time during debriefing (69.5% vs 49.0%, P < 0.0001).
- Trainees in the AI group reported higher post-training scores for training content (4.7 ± 0.5 vs 3.6 ± 1.0, P = 0.0076), satisfaction (4.9 ± 0.3 vs 3.5 ± 0.8, P < 0.0001), and procedural confidence (5.0 ± 0.0 vs 4.5 ± 0.7, P = 0.0216).
