Key points are not available for this paper at this time.
OBJECTIVE: The objective of our study was to evaluate the performance of liver CT in the diagnosis of esophageal varices in patients with cirrhosis and to determine whether thin-section multiplanar reconstructions (MPRs) improve accuracy. MATERIALS AND METHODS: We identified 109 patients with cirrhosis who underwent endoscopy within 10 weeks after dual-phase liver MDCT supplemented with thin-section axial and coronal portal venous phase reconstructions. Two blinded radiologists independently evaluated each CT examination for the presence and sizes of varices using standard 5-mm axial versus 1- to 3-mm multiplanar images in separate sessions. Sensitivity, specificity, and predictive value calculations and receiver operating characteristic analysis were performed using endoscopy as the reference standard. Interobserver variability and correlation of CT size to variceal grade were assessed. RESULTS: Twenty-six cases of high-risk esophageal varices were identified; all except two were detected on CT by one of the readers on standard 5-mm images. For both readers, sensitivity and negative predictive value (NPV) for the discrimination of high-risk varices using a criterion of 2 mm or greater were nearly the same for the standard 5-mm images versus the 1- to 3-mm multiplanar images (sensitivity and NPV: reader 1, 96% and 98% vs 96% and 99%; reader 2, and 89% and 95% vs 89% and 96%, respectively). Standard 5-mm images yielded a lower specificity for high-risk esophageal varices than the thin-section multiplanar images, and this difference was statistically significant for reader 2. Substantial interobserver agreement was noted for both esophageal varices detection and size measurements. CONCLUSION: Standard liver CT is sensitive for the detection of high-risk varices and deserves further investigation as a potential cost-effective screening tool for the evaluation of patients with cirrhosis. The addition of 1- to 3-mm MPRs may increase specificity for risk stratification based on size measurements.
Yu et al. (Tue,) studied this question.