Integrated diagnosis of submassive pulmonary embolism using computed tomography angiography, echocardiography, and artificial intelligence

Journal of Vascular Surgery,2026

Abstract

Objectives

Early intervention in submassive pulmonary embolism (PE) has been shown to improve long-term cardiopulmonary outcomes compared with anticoagulation alone. Submassive PEs are diagnosed by documentation of a right-to-left ventricular (RV/LV) ratio of >0.9, indicative of right heart strain (RHS), and are associated with adverse clinical outcomes. Although often used interchangeably to guide the treatment of submassive PE, limited data exist comparing measurement of RV/LV ratio by computed tomography angiography (CTA) and transthoracic echocardiography (TTE). We also examined the role of artificial intelligence (AI) in the early detection of submassive PEs.

Methods

A single-institution retrospective review of AI-detected PE (Viz.ai) on CTA of the chest was performed over 1 year. Detection of PE on CTA by AI (CTA/AI) activated the PE response team. Submassive PE was defined by a RV/LV ratio of >0.9. A TTE was performed in all patients with submassive PE on CTA confirmed by radiology (CTA/RAD). In addition, some patients had TTE based on their clinical condition despite having a RV/LV ratio of <0.9 by CTA. Using χ2 analysis, we compared ability to detect submassive PE by CTA/RAD with TTE.

Results

Over 1 year, 201 Viz.ai activations for suspected PE on CTA of the chest were reviewed (n = 112 male [56%], n = 89 female [44%]; mean age, 72 ± 17 years). PE was detected in 121 patients on CTA/RAD, and 105 patients had an RV/LV ratio calculated on both CTA/RAD and TTE during evaluation for submassive PE. Forty-seven of the 105 patients (45%) had RHS on CTA/RAD, but no RHS on TTE. When comparing CTA/RAD with TTE, the average RV/LV ratio was 0.32 ± 0.34 (P > .0001) higher than on CTA/RAD. When compared with CTA/RAD, CTA/AI had a positive predictive value of 0.6 for the detection of any PE. Using TTE as the standard, CTA/RAD had a positive predictive value of 0.44 and a negative predictive value of 0.9 for the detection of submassive PE.

Conclusions

Our study suggests the need for further improvements in CTA/AI PE detection and that CTA/RAD is suboptimal in the evaluation of submassive PE in isolation. We also noted that CTA/RAD significantly overestimates the RV/LV ratio and thereby RHS compared with TTE. Basing the need for intervention solely on CTA/RAD may subject patients unnecessarily to the risks associated with intervention for submassive PE. Our data suggest the importance of including both CTA and TTE in the PE response team algorithm, and that these imaging modalities should not be used interchangeably, but synergistically to guide treatment for submassive PE.

Read the full study.