A new clinical trial shows that AI can guide untrained operators to perform deep vein thrombosis scans, but the real value lies in filtering the patient queue rather than replacing human specialists.

Can a medical novice with a handheld probe safely rule out a deadly blood clot? For years, diagnosing deep vein thrombosis (DVT) has required a trained sonographer. This bottleneck often leaves patients waiting hours in emergency rooms for a simple scan. This new trial challenges the assumption that we must wait for specialists to get a reliable diagnostic image.

The findings reveal that AI-guided imaging is not a magic diagnostic shortcut, but a highly effective triage filter. By letting non-experts rule out low-risk patients, hospitals can dramatically shorten waiting lines. This shifts the focus of clinical AI from autonomous diagnosis to smart queue management.

In this multicenter, double-blinded study, researchers evaluated the ThinkSono Guidance system. Out of 634 recruited subjects, they analyzed 594 patients, capturing 67 DVTs across 700 scans. Non-ultrasound-trained operators performed the scans, which were then sent to remote clinicians for review.

The triage bottleneck shifts

The system proved remarkably fast, though it has clear boundaries. The median scan and review time was just 7.57 minutes. This speed could alter how emergency departments manage suspected clot cases after hours.

Key metrics from the trial include:

• Diagnostic image quality was achieved in 86.83% of the AI-guided scans.
• Triage sensitivity reached 92.86%, ensuring most clots were caught.
• Triage specificity was low at 39.12%, meaning many healthy patients still flagged as needing follow-ups.
• Prioritization specificity hit 97.96%, allowing clinicians to safely rule out DVT in some cases.
• Standard of care ultrasounds could be completely avoided in 35.32% of patients.

The low triage specificity of 39.12% is the critical detail here. It means the system over-refers patients for standard scans. However, because the prioritization specificity is so high, clinicians can confidently send more than a third of patients home without a formal radiology consult. This aligns with earlier feasibility work, such as a 2024 feasibility study on machine learning for DVT, which highlighted the potential of shifting initial assessments to primary care. Another 2025 study in NEJM AI similarly proved that AI-driven detection can reliably spot proximal DVTs.

Real-world clinical limits

This is not a hands-off solution. The system still requires a remote clinician to interpret the images, meaning it does not eliminate human labor. Furthermore, more than 13% of the scans failed to reach diagnostic quality, which could cause anxiety or delays if not managed.

But as a tool to relieve overburdened emergency departments, this approach is highly practical. It proves that AI does not need to be perfect to be useful. It just needs to safely clear the easy cases out of the way.

Read the full study in medRxiv.