In this internal retrospective evaluation of 100 patients with suspected PAH, a multimodal AI model integrating chest CT and echocardiography showed improved overall discrimination versus CT or echocardiography alone (against right heart catheterization), and should be considered an auxiliary, noninvasive tool requiring external prospective validation.
The multimodal AI model achieved high diagnostic performance for PAH against right heart catheterization as the reference standard.
Chest CT alone had lower diagnostic performance than the multimodal AI model.
Echocardiography alone had the lowest diagnostic performance of the three approaches.
The study used a small single-cohort retrospective design with all three tests performed within a short interval.
Exploratory subgroup analyses were performed for borderline pulmonary artery diameter and connective tissue disease.
The authors conclude the multimodal approach is an auxiliary noninvasive tool needing external prospective validation.
Pulmonary arterial hypertension (PAH) is a serious condition in which blood pressure in the lung arteries is too high, straining the right side of the heart. The standard way to confirm it is right heart catheterization, an invasive procedure. Chest CT and heart ultrasound are noninvasive but each misses some cases or gives false alarms when used alone. This study looked back at 100 patients with suspected PAH who had a chest CT, a heart ultrasound, and catheterization within two weeks of each other. Researchers built an AI model that combined information from both images and compared it with each test alone, using catheterization as the reference. The AI correctly identified PAH in about 92% of those who had it and correctly ruled it out in about 89% of those who did not, with overall accuracy of 91%. CT alone had 83% accuracy and ultrasound alone had 79%. Statistical tests showed the AI's classifications differed significantly from both. This research suggests that combining imaging types with AI could help support noninvasive PAH diagnosis. However, the study was small, retrospective, and tested the model only on internal data, so the results may not hold in other hospitals or patient groups. The authors state that the tool should be considered only an auxiliary aid and needs external, prospective testing before it could be relied on in practice.
Sun Z, Liu F. (2026). Multimodal Artificial Intelligence Model Integrating Chest CT and Cardiac Ultrasound for the Auxiliary Diagnosis of Pulmonary Arterial Hypertension.. Journal of visualized experiments : JoVE. https://doi.org/10.3791/72551