Assessment of Hepatic Vascular Network Connectivity with Automated Graph Analysis of Dynamic Contrast-enhanced US to Evaluate Portal Hypertension in Patients with Cirrhosis: A Pilot Study

Purpose: To test whether graph analysis of vascular images obtained with hepatic dynamic contrast material-enhanced (DCE) ultrasonography (US) allows calculation of the degree of organization of the liver circulation and whether graph properties are correlated to the severity of portal hypertension. Materials and Methods: Institutional review board approval and written informed consent were obtained. Fifteen patients with liver cirrhosis (nine men; mean age 6 standard deviation, 55 years 6 8) who underwent DCE US and hepatic venous pressure gradient (HVPG) measurement and four healthy subjects (two men and two women; mean age, 34 years 6 4) were included between January 2007 and December 2008. Individual graph models ("vascular connectomes") were computed on the basis of time series analysis of video sequences of DCE US examinations (conducted with the disruption-reperfusion technique). Graph analysis was performed, and the clustering coefficient C was calculated. Correlations between clustering coefficient and HVPG were assessed. Results: Healthy subjects had a high clustering coefficient of vascular connectome (C = 0.4447; interquartile range [IQR], 0.3864-0.4679), suggesting a highly organized hepatic vascular network. Conversely, patients with cirrhosis showed a low clustering coefficient, indicating disruption of normal anatomy (C = 0.0288; IQR, 0.0157-0.0861; P = .001 vs healthy subjects). The clustering coefficient decreased as HVPG increased, with a clustering coefficient of 0.0237 (IQR, 0.0066-0.0378) in patients with HVPG of at least 10 mm Hg versus 0.1180 (IQR, 0.0987-0.1414) in those with HVPG of less than 10 mm Hg (P = .006). The correlation between the best model derived from the distribution of the clustering coefficient (10 bins) of vascular connectome and HVPG had a Pearson correlation of 0.977 (root mean squared error, 1.57 mm Hg; P < .0001). Conclusion: This pilot study demonstrates that graph modeling of vascular connectivity based on video processing of liver DCE US examinations and subsequent graph analysis enable calculation of personalized parameters that reflect the degree of organization of the hepatic microvascular network and are correlated to the severity of portal hypertension in cirrhosis. (C) RSNA, 2015