aceExplainable machine learning for early prediction and anatomical classification of pulmonary embolism in the emergency department
Pulmonary thromboembolism (PTE) is a life‑threatening condition that requires prompt and accurate evaluation in the emergency department (ED). Standardized clinical scoring systems, including the Wells and revised Geneva scores, form the cornerstone of initial risk stratification but have limited specificity, leading to unnecessary D‑dimer testing and frequent overuse of CT pulmonary angiography (CTPA). This study aimed to develop explainable machine‑learning (XML) models as a complementary decision‑support laye…
United States Navy Medical News Letter Vol. 51, No. 12, 21 June 1968 by U.S. Navy. Bureau of Medicine and Surgery. Public domain
In a 2022-2024 study of 472 emergency department patients with suspected pulmonary thromboembolism across three Mashhad University of Medical Sciences centers, researchers developed explainable machine-learning models intended to operate after Wells/Geneva triage. Using CTPA as reference, Extra Trees achieved accuracy 0.82, sensitivity 0.69, specificity 0.86 and AUC 0.83 for PTE prediction, and AUC 0.77 for central and 0.67 for peripheral emboli for anatomical classification.
The work matters because standard scores have limited specificity and drive overuse of D-dimer and CTPA, and a transparent model could help reduce avoidable imaging while providing early non-imaging indication of clot location. Uncertainty remains about generalizability beyond this intermediate- to high-probability cohort where only 24% were CTPA-positive, and about clinical impact given the 63% anatomical classification accuracy reported by July 2026.
- Study collected clinical and paraclinical data from 472 ED patients with suspected PTE across three centers of Mashhad University of Medical Sciences (2022-2024).
- CTPA served as reference standard; in this intermediate- to high-probability cohort only 24% were confirmed positive on CTPA despite all being classified as PTE-suspect by Wells/Geneva.
- Models included classical ML, ensemble algorithms such as Extra Trees, and hybrid stacking pipelines with SHAP values for explainability.
- Among PTE-positive cases, model achieved AUC of 0.77 for central emboli and 0.67 for peripheral emboli.
Explainable ML models used after Wells/Geneva triage improved early PTE prediction in ED patients, with Extra Trees reaching 0.82 accuracy and 0.83 AUC.
The rundown
Researchers built explainable ML models as a complementary layer after Wells and revised Geneva scoring, using structured forms aligned with those criteria and CTPA as reference. The design aimed to support D-dimer decisions in low-risk patients and refine risk estimation in intermediate- and high-risk patients.
In 472 patients from three Mashhad University centers between 2022-2024, Extra Trees performed best for PTE prediction. For clot location, discrimination was higher for central than peripheral emboli, with overall anatomical classification accuracy reported at 63%.
Sources
- Peer-reviewedBMC Emergency Medicine2026-07-17
How should this claim be treated?
The debate