Authors' objectives: To identify and summarise evidence that addresses the following question: What is the clinical and cost effectiveness of artificial intelligence (AI) assisted ultrasound for the detection and diagnosis of deep vein thrombosis (DVT)?

Authors' results and conclusions: The evidence included in this review suggests that the diagnostic accuracy of AI-assisted ultrasound is variable. On average, one study found ThinkSono Guidance to be quicker than the gold standard of care. Three studies report that most study participants were identified as low risk in their cohorts, suggesting the potential avoidance of a duplex ultrasound scan. HTW researchers did not identify outcomes relating to quality of life or avoidance of interim therapeutic anticoagulation. The final price of ThinkSono Guidance was not available at the time of our analysis. We estimated that at the expected price range (£113 to £208 per scan), the intervention strategy would be dominated by usual care, because it was overall more expensive and less effective.

Authors' recommendations: More evidence is needed on ThinkSono Guidance for the detection and diagnosis of deep vein thrombosis.

Authors' methods: The Evidence Appraisal Report is based on a literature search (strategy available on request) for published clinical and economic evidence on the health technology of interest. It is not a full systematic review but aims to identify the best available evidence on the health technology of interest. Researchers critically evaluate and synthesise this evidence. We include the following clinical evidence in order of priority: systematic reviews; randomised trials; non-randomised trials. We only include evidence for “lower priority” evidence where outcomes are not reported by a “higher priority” source. We conducted meta-analysis of diagnostic test accuracy by fitting the bivariate model to sensitivity/specificity points using a linear mixed model using the reitsma() function in the mada package in R version 4.5.1. Heterogeneity was assessed visually by looking at the receiver operating curve (ROC) plot of sensitivity and specificity. We also search for economic evaluations or original research that can form the basis of an assessment of costs/cost comparison. We carry out various levels of economic evaluation, according to the evidence that is available to inform this.

Authors' identified further research: The Panel agreed that more robust evidence is needed on real-world system outcomes, including impacts on waiting times, the use of interim anticoagulation, and quality of life. The Panel agreed further evidence should be generated to reduce uncertainty around diagnostic accuracy and outcomes of the proposed pathway compared with usual care.

Project Status: Completed

Year Published: 2026

URL for published report: https://healthtechnology.wales/reports-guidance/artificial-intelligence-assisted-ultrasound/

English language abstract: An English language summary is available

Publication Type: Rapid Review

Country: Wales, United Kingdom

Organisation Name: Health Technology Wales

Contact Address: c/o Digital Health Care Wales, 21 Cowbridge Road East Cardiff CF11 9AD

Contact Name: Susan Myles, PhD

Contact Email: healthtechnology@wales.nhs.uk