Diagnostic accuracy of an automated artificial intelligence derived right ventricular to left ventricular diameter ratio tool on CT pulmonary angiography to predict pulmonary hypertension at right heart catheterisation

• Published in: Clinical radiology Vol. 77; no. 7; pp. e500 - e508
• Main Authors: Charters, P.F.P., Rossdale, J., Brown, W., Burnett, T.A., Komber, H.M.E.I., Thompson, C., Robinson, G., MacKenzie Ross, R., Suntharalingam, J., Rodrigues, J.C.L.
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 01-07-2022
• Subjects: Angiography - methods; Artificial Intelligence; Cardiac Catheterization; Heart Ventricles - diagnostic imaging; Humans; Hypertension, Pulmonary - diagnostic imaging; Pulmonary Artery - diagnostic imaging; Retrospective Studies; Tomography, X-Ray Computed - methods
• ISSN: 0009-9260; 1365-229X
• DOI: 10.1016/j.crad.2022.03.009

To determine the diagnostic accuracy of an automated artificial intelligence derived right ventricle/left ventricle diameter ratio (RV/LV) computed tomography pulmonary angiography (CTPA) analysis tool to detect pulmonary hypertension (PH) in patients with suspected PH referred to a specialist centre.
The present study was a retrospective analysis of a prospectively maintained database of 202 consecutive patients with suspected PH, who underwent CTPA within 12 months of right heart catheterisation (RHC). Automated ventricular segmentation and RV/LV calculation (Imbio LLC, Minneapolis, MN, USA) was undertaken on the CTPA images. PH diagnosis was made using the RHC reference standard.
The automated RV/LV correlated more strongly with RHC metrics than main pulmonary artery (MPA) diameter and MPA to ascending aorta diameter ratio (MPA/AA) measured manually (mean pulmonary arterial pressure [mPAP] r=0.535, R2 = 0.287 p<0.001; pulmonary vascular resistance [PVR] r=0.607, R2 = 0.369 p<0.001). In the derivation cohort (n=100), the area under the receiver-operating curve for automated RV/LV discriminating PH was 0.752 (95% confidence interval [CI] 0.677–0.827, p<0.001). Using an optimised Youden's Index of ≥1.12 classified from derivation, automated RV/LV ratio analysis was more sensitive for the detection of PH with higher positive predictive value (PPV) when compared with manual MPA and MPA/AA in the validation cohort (n=102). Automated RV/LV compromise (1.12) and specific (1.335) thresholds were strongly predictive of mortality (log-rank 7.401, p=0.007 and log-rank 16.075, p<0.001 respectively).
In suspected PH, automated RV/LV diameter thresholds have high sensitivity for PH, outperform manual MPA and MPA/AA and can predict survival.
•Automated RV/LV analysis is feasible in real-world suspected pulmonary hypertension.•Automated RV/LV thresholds perform favorably compared with manual metrics on CT in PH.•Automated RV/LV analysis can predict survival in patients with suspected PH.