COIOS
Weighing data and models Item new preprint

AI-guided ultrasound by untrained operators detected 93% of proximal DVTs on remote review

In a multicentre prospective study of 594 patients with 700 scans, ultrasounds acquired by non-ultrasound-trained operators under AI guidance and read remotely reached diagnostic quality in 86.8% of scans, with triage sensitivity of 92.9% for proximal deep vein thrombosis and specificity of 39.1%.

A prospective, double-blinded, non-randomised multicentre study compared compression ultrasounds of the proximal lower limb acquired by non-ultrasound-trained operators using an AI guidance system (ThinkSono Guidance) with standard-of-care ultrasound in 594 analysed patients, covering 700 scans and 67 proximal DVTs. Clinician-reviewed AI-guided scans were of diagnostic quality in 86.8% of cases, with triage sensitivity 92.9%, triage specificity 39.1% and prioritisation specificity 98.0%. Median combined scan and review time was 7.57 minutes, and standard-of-care ultrasound was reported as avoidable in 35.3% of patients.

Why it is interesting: The AI guidance changes who can perform the ultrasound rather than who reads it, with accuracy reported against a concurrent reference standard.

Source
medRxiv, 27 September 2026
DOI
10.64898/2026.07.16.26356882
Type
Preprint
Design
Multicentre, double-blinded, prospective non-randomised diagnostic accuracy study; 594 patients analysed, 700 scans, 67 proximal DVTs
Verdict
New finding
Driver
AI in medicine