Machine Learning with Contrast-Enhanced CT Can Help Identify Pulmonary Hypertension

An automated approach can be useful with patients who have indeterminate transthoracic echocardiograms.

Using automated machine learning to calculate the right ventricle/left ventricle (RV/LV) ratio can help radiologists identify patients who have pulmonary hypertension (PH) and would benefit from being referred to a specialist.

In a presentation during the 2021 European Congress of Radiology annual meeting, Pia Charters from Royal United Hospital Bath NHS Foundation Trust outlined the results of a study that examined the efficacy and feasibility of using this automated analysis to better predict PH in patients who receive indeterminate results with transthoracic echocardiogram (TTE).

For more coverage of ECR 2021, click here.

While TTE is the typical screening method used to detect PH, it has poor agreement with right heart catheter (RHC) gold standard, she noted. Consequently, there is a need for a better assess patients who receive unclear TTE results.

Charters’ team conducted a retrospective study with 241 patients who underwent RHC between June 2017 and March 2018. They also had either CTPA or TTE within the previous 12 months. The team used measurements decided by both the European Society of Cardiology (ESC) and the World Symposium of PH (WSPH) to identify patients with PH. Individuals were considered to have PH by the RHC gold standard if their mPAP was greater than or equal to 25 mmHg (ESC) or between 20-24 mmHg with pulmonary vascular resistance greater than 3 (WSPH).

Based on their analysis, the team determined that 179 patients met the ESR definition, and 181 satisfied the WSPH definition. Patients with PH tended to be older with a median age of 67 and female, Charters said.

In addition, the team identified a mean automated RV/LV for ESC and WSPH PH of 1.37 and a mean automated RV/LV for mPAP <20 mmHg of 1.09. There was a significant, strong relationship between the RV/LV ratio and mPAP and the RV/LV ratio and PVR, she said.

Their results also revealed that automated RV/LV had high diagnostic accuracy for detecting PH with the area under the curve of 0.796. Also, RV/LV >0.935 was more sensitive in detecting both ESC and WSPH PH than TTE or a manual pulmonary ratio.

Overall, Charters said, the automated RV/LV threshold has high sensitivity for both ESC- and WSPH-defined PH, and it offer diagnostic utility when patient’s return with an indeterminate TTE score. Not only does the ratio correlate with mPAP and PVR, but the analysis can also be used to increase or decrease the pre-test probability of those patients. There is also the possibility that it could change the impact of screening.

“There is the potential to screen all patients undergoing CTPA for PH, identifying them radiologically as opposed to clinically, with emphasis on early detection and earlier intervention,” she said.

For more coverage based on industry expert insights and research, subscribe to the Diagnostic Imaging e-Newsletter here.