AI-based pulmonary artery to ascending aorta ratio on non-contrast CT for pulmonary hypertension: diameter vs. volume assessment.
Authors
Affiliations (7)
Affiliations (7)
- School of Medicine and Population Health, The University of Sheffield, Sheffield S10 2TN, UK.
- Radiological Sciences Department, College of Applied Medical Science, King Saud bin Abdulaziz University for Health Science, Riyadh 11426, Saudi Arabia.
- King Abdullah International Medical Research Centre (KAIMRC), Riyadh 11426, Saudi Arabia.
- Insigneo Institute, Faculty of Engineering, The University of Sheffield, Sheffield, UK.
- Department of Clinical Radiology, Sheffield Teaching Hospitals, Sheffield, UK.
- National Institute for Health and Care Research, Sheffield Biomedical Research Centre, Sheffield, UK.
- Sheffield Pulmonary Vascular Disease Unit, Sheffield Teaching Hospitals NHS Trust, Sheffield, UK.
Abstract
To assess the diagnostic accuracy of a deep learning (DL) model for quantifying the pulmonary artery (PA) and ascending aorta (AAo) diameters and volumes on non-contrast computed tomography (CT) scans for detecting pulmonary hypertension (PH). The PA and AAo were segmented using a validated DL model on non-contrast CT scans. PA/AAo ratios were quantified using diameter and volume measurements. Testing was performed using two independent patient cohorts. A first cohort (<i>n</i> = 97) was used to compare DL-derived and manual diameter-based PA/AAo ratios and to evaluate their diagnostic accuracy against right heart catheterization (RHC). A second cohort (<i>n</i> = 425; 385 internal and 40 external patients) was then used to compare the diagnostic performance of the DL-derived diameter- and volume-based PA/AAo ratios using RHC as the reference standard. In the first cohort, 97 patients were included (mean age 63 ± 13 years, 50% female, and 62 patients with PH). Agreement with the diameter-based approach was good [Observer 1: intraclass correlation coefficient (ICC) = 0.79 (0.70-0.85); Observer 2: ICC = 0.79 (0.70-0.86)]. Agreement with volume-based PA/AAo ratio was moderate [Observer 1: ICC = 0.56 (0.41-0.68); Observer 2: ICC = 0.55 (0.39-0.67)]. The DL diameter-based (area under the curve (AUC) = 0.83) and volume-based (AUC = 0.84) PA/AAo ratio outperformed the manual measurements by Observer 1 (AUC=0.72) and Observer 2 (AUC = 0.74). The second cohort comprised 425 patients (385 internal and 40 external), with a mean age of 65 ± 12 years, 60% female, and 353 patients with PH. The volume-based approach achieved superior diagnostic accuracy compared with the diameter-based approach (AUC 0.85 vs. 0.73), with higher sensitivity (66% vs. 41%) and specificity (89% vs. 82%). The PA/AAo volume ratio on non-contrast CT demonstrates high diagnostic accuracy for detecting PH and could assist with opportunistic detection of PH in patients undergoing routine chest imaging.