AI nerve tracking did not teach novices to find the nerve

Design

prospective randomised trial, assessment with AI disabled

Population

80 ultrasound novices (60 medical students, 20 anaesthesiology residents); 78 completed

Primary outcome

first-attempt independent identification of median nerve and supraclavicular brachial plexus

Effect

median nerve 65.8% vs 57.5% (RR 1.14, 95% CI 0.81 to 1.62); brachial plexus 42.1% vs 35.0% (RR 1.20, 0.68 to 2.11)

Eighty ultrasound novices — 60 medical students and 20 anaesthesiology residents — were randomised to standard B-mode training or to training with a real-time AI nerve tracking system, then assessed with the AI switched off. The primary outcome was first-attempt independent identification of the median nerve and the supraclavicular brachial plexus.

Neither improved. Median nerve identification succeeded in 57.5 per cent of the control group and 65.8 per cent with AI (RR 1.14, 95% CI 0.81 to 1.62, P = .452); brachial plexus in 35.0 per cent and 42.1 per cent (RR 1.20, 0.68 to 2.11, P = .519). After further AI-assisted training the AI group reported more confidence in identifying the brachial plexus (4.7 ± 0.5 vs 4.4 ± 0.9, P = .046) — with no corresponding improvement in whether they could actually do it.

That gap between confidence and performance is the finding worth carrying. The authors are careful to say the trial does not establish anything about long-term retention, transfer to unfamiliar anatomy, or clinical procedural performance, and it does not test AI as an intraoperative aid — only as a teaching tool, after 10 minutes of hands-on practice.

• Do not substitute AI overlay for supervised scanning practice in teaching regional anaesthesia
• Watch for confidence rising faster than competence in trainees using assistive tools
• Assess trainees with the assistance disabled, not with it running
• Note this tested teaching, not intraoperative use — the clinical question is still open

Why it matters

Assistive tools are being bought as training aids on the assumption that using them teaches the skill.

Don't overread it

This tested a brief training intervention in novices, not intraoperative AI assistance or block success.

The statistics, in plain English

Both confidence intervals comfortably include 1, so this is a null result rather than evidence of harm — though with 39 per arm the trial could not have detected a small benefit. The confidence finding at P = .046 is one of several secondary outcomes and sits right at the conventional threshold.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.