Liver surface nodularity quantification from routine ct images as a biomarker for detection and evaluation of cirrhosis

Purpose: To determine the accuracy, reproducibility, and intra-and interobserver agreement of a computer-based quantitative method to measure liver surface nodularity (LSN) from routine computed tomographic (CT) images as a biomarker for detection and evaluation of cirrhosis. Materials and Methods: For this institutional review board-approved HIPAA-compliant retrospective study, adult patients with healthy livers (n = 24) or various stages of hepatitis C virus-induced chronic liver disease (n = 70) with routine nonenhanced and portal venous phase contrast agent-enhanced liver CT imaging with thick-section (5.0 mm) and thin-section (1.25-1.50 mm) axial images obtained between January 1, 2006, and March 31, 2011, were identified from the electronic medical records. A computer algorithm was developed to measure LSN and derive a score. LSN scores, splenic volume, and the ratio of left lateral segment (LLS) to total liver volume (TLV) were measured from the same multiphasic liver CT examinations. Accuracy for differentiating cirrhotic from noncirrhotic livers was assessed by area under the receiver operating characteristic curve. Intra-and interobserver agreement was assessed by intraclass correlation coefficient. Results: Median LSN scores from nonenhanced thick-section CT images in cirrhotic livers (3.16; 56 livers) were significantly higher than in noncirrhotic livers (2.11; 38 livers; P , .001). LSN scores from the four CT imaging types (94 patients for each type) were very strongly correlated (range of Spearman r, 0.929-0.960). LSN scores from portal venous phase contrast-enhanced thick-section CT images had significantly higher accuracy (area under the receiver operating characteristic curve, 0.929) than splenic volume (area under the receiver operating characteristic curve, 0.835) or LLS-to-TLV ratio measurements (area under the receiver operating characteristic curve, 0.753) for differentiating cirrhotic from noncirrhotic livers (P = .038 and .003, respectively; n = 94). Intra-and interobserver agreements that used nonenhanced thick CT images were very good (intraclass correlation coefficient, 0.963 and 0.899, respectively). Conclusion: Quantitative measurement of LSN on routine CT images accurately differentiated cirrhotic from noncirrhotic livers and was highly reproducible.