Hemodynamic measurement in pulmonary hypertension: a challenge for automated CT analysis

<bold>AIM:</bold> Pulmonary hypertension (PH) is defined by an increased pulmonary arterial pressure that can emerge from various parts of the pulmonary vascular bed. Its diagnosis is based on right heart catheterization (RHC). The identification of new biomarkers in PH is currently needed. <bold>METHODS:</bold> We compared CT scans evaluation using AI-based vascular quantification with invasive hemodynamic assessment. The dataset consisted of patients from a retrospective cohort collected at the University Hospital of Liege, including adults who underwent a RHC and a CT scan within 6 months. Vascular quantification was performed with a dual approach consisting in segmentation of the vascular bed using deep learning (Radiomics.bio, Belgium) and geometrical analysis of the full pulmonary vasculature. Patients were then separated between pre- and post-capillary PH according to the hemodynamic definition. RESULTS We analyzed 60 patients including 17 with post-capillary PH and 43 pre-capillary PH (18 PAH, 10 Gr3 and 15 Gr4). A Mann-Whitney U-test identified a significant difference (p<0.05) between normalized arterial volume for vessels with radius within 3 and 4mm considering post-capillary PH vs PAH and considering post-capillary PH vs other groups. The mean pulmonary artery (PA) median and maximal normalized diameters were also significantly different between Gr2 and other groups as well as the left and right PA maximal normalized diameters. <bold>CONCLUSION:</bold> Some imaging biomarkers showed a statistically significant difference between group 2 PH and the other groups. These results suggest that AI-based quantification tools could help for a better classification and characterization of PH as a first step before invasive evaluation.