In a new study from UHN, researchers used machine learning models to predict the severity of chronic thromboembolic pulmonary hypertension (CTEPH)—a rare but treatable condition where there is abnormally high blood pressure in the lungs due to old blood clots and scar tissue. Their findings could help clinicians better assess risk and guide treatment.
CTEPH can develop after a pulmonary embolism (PE)—a blood clot that travels to the lungs. The condition can be cured with surgery, known as pulmonary endarterectomy, which removes the old clots and scar tissue.
To decide if a patient is a good candidate for this surgery, clinicians rely on CT Pulmonary Angiogram (CTPA) scans to assess the extent and location of the blockages. However, current methods for analyzing CTPA results, such as scoring blood vessel obstruction, are not reliable for predicting the severity of the patient’s disease.
To address these gaps and re-examine the link between PE and CTEPH, researchers tested whether machine learning, a form of AI that finds patterns in data, could use detailed blood clot data from CTPA scans to find better links to severe lung high blood pressure (pulmonary hypertension) observed in patients with CTEPH.
The team studied 184 patients with CTEPH who had surgery at UHN between 2017 and 2022, led by Dr. Marc de Perrot, Senior Scientist and Thoracic Surgeon and Dr. Laura Donahoe, Thoracic Surgeon. They found that about 22% of patients had severe pulmonary hypertension. As in earlier studies, the amount of clotting seen on scans alone did not predict how severe the pulmonary hypertension was.
Instead, the most reliable indicator was the right-to-left ventricle ratio—a simple and easy-to-use measurement comparing the size of the heart’s two main pumping chambers. A ratio above 1.4 was strongly associated with severe disease.
The machine learning models tested were also able to identify severe cases of CTEPH by combining this heart measurement with factors such as patient age, sex, and blood clot details from the CTPA scans. This shows that multiple factors influence how serious the condition becomes.
For radiologists, clinicians, and patients, these results provide a new reference point to help identify patients with severe disease and highlight how combining imaging data and machine learning can support better care.
Dr. Micah Grubert Van Iderstine, a former medical student at the University of Manitoba and current resident in the UBC Diagnostic Radiology Residency Program, is the first author of the study.
Dr. Micheal McInnis is a Clinician Investigator at UHN and Assistant Professor in the Department of Medical Imaging at the University of Toronto. He is the corresponding author of this study.
This work was supported by UHN Foundation.
Dr. Micheal McInnis receives speaker fees from Boehringer Ingelheim and AstraZeneca (ongoing) and formerly sat on an advisory board for Boehringer Ingelheim and AstraZeneca (concluded).
Grubert Van Iderstine M, Kim S, Karur GR, Granton J, de Perrot M, McIntosh C, McInnis M. Utility of machine learning for predicting severe chronic thromboembolic pulmonary hypertension based on CT metrics in a surgical cohort. Eur Radiol. 2025 Aug 23. doi: 10.1007/s00330-025-11972-9. Epub ahead of print.