A UK team found that software can measure a key heart‑scan signal with the same accuracy as clinicians—potentially speeding up screening for pulmonary hypertension (PH), a serious form of high blood pressure in the lungs.  

Why this matters

Pulmonary hypertension is often missed for years because early symptoms—shortness of breath, fatigue—mimic other conditions. Echocardiography (a standard heart ultrasound) is the first‑line test, but some readings are time‑consuming and vary by operator. The new study tested whether AI could automatically read one of the most important markers, the tricuspid regurgitation jet velocity (TRJV), to help clinicians spot PH sooner.  

What the researchers did

Scientists at the University of Sheffield analyzed 1,031 adults evaluated for suspected PH. Every patient had both an echocardiogram and the definitive—but invasive—right‑heart catheterization. They compared the AI tool (US2.AI, which has U.S. FDA clearance) against manual measurements by clinicians, using catheterization results as the gold standard. The study population and design are shown in the flow chart on page 5.  

What they found (in plain language)

• Same overall accuracy as humans. The AI’s ability to identify PH from TRJV matched manual readings (AUC 0.88 vs. 0.88), as illustrated in the graphical abstract on page 1 and detailed in Table 2 (page 7).  
• High agreement, minimal bias. AI and human TRJV values were almost interchangeable (intraclass correlation 0.94), with a tiny average difference on the Bland–Altman plot (page 6).  
• More scans successfully measured. The AI produced a TRJV result in 87% of echocardiograms vs. 80% for manual reads—important in busy clinics.  
• Clear trade‑offs at standard cut‑offs. Using the higher TRJV threshold (3.4 m/s) gave strong rule‑in specificity (~92–93%) but lower sensitivity (~67–68%); the lower threshold (2.8 m/s) boosted sensitivity (~89%) while dropping specificity (~60–63%). See Table 2 (page 7).  
• Blood‑pressure estimate aligned with the catheter test. Systolic pulmonary artery pressure calculated from the AI’s TRJV closely matched invasive measurements (agreement ICC 0.83–0.85).  

What it means for patients

AI that reads routine heart ultrasounds could standardize measurements, reduce delays, and help teams decide sooner who needs specialist testing—without replacing right‑heart catheterization, which remains the diagnostic standard. In clinics with fewer expert readers, a reliable AI assist could be especially helpful.  

Caveats

This was a retrospective analysis of patients already referred for suspected PH, so PH was common in the group; real‑world performance in general clinics may differ. The tool was originally trained on heart‑failure scans, and automated readings for some other PH‑related echo signs were less successful (e.g., right‑atrial area in 43% of cases), underscoring the need for prospective trials across diverse populations. Funding included industry support, and two co‑authors were employees of the tool’s developer.  

Bottom line

An FDA‑cleared AI can read a key heart‑ultrasound signal as accurately as specialists and more consistently, pointing to faster, more dependable PH screening. It’s a promising add‑on for clinics—but it should support, not replace, expert judgment and confirmatory testing.  

Source: Salehi M, Alabed S, Sharkey M, et al. “Artificial intelligence‑based echocardiography assessment to detect pulmonary hypertension,” ERJ Open Research (2025).  

Editor’s note: This article is for information only and is not a substitute for professional medical advice.