Diagnostic accuracy of imaging for liver cirrhosis compared to histologically proven liver cirrhosis. A multicenter collaborative study

Abstract

Objective: To evaluate the diagnostic accuracy of liver cirrhosis by imaging modalities, including CT, MRI and US, compared to results obtained from histopathological diagnoses of resected specimens.

Materials and methods: CT, MRI and US examinations of 142 patients with chronic liver disease who underwent surgery for complicated hepatocellular carcinoma (<3 cm in diameter) in 10 institutions were blindly reviewed in a multicenter study by three radiologists experienced in CT, MRI and US. The images were evaluated for five imaging parameters (irregular or nodular liver surface, blunt liver edge, liver parenchymal abnormalities, liver morphological changes and manifestations of portal hypertension) using a severity scale. The diagnostic imaging impression score was also calculated. Patients were histologically classified into chronic hepatitis (CH; n = 54), liver cirrhosis (LC; n = 71) and pre-cirrhosis (P-LC; n = 17) by three pathologists, independently, who reviewed the resected liver specimens. The results of the three imaging methods were compared to those from histological diagnoses, and a multivariate analysis (stepwise forward logistic regression analysis) was performed to identify independent predictive signs of cirrhosis. The diagnostic efficacies for LC and early cirrhosis were also compared among CT, MRI and US using a receiver-operating characteristic (ROC) curve analysis.

Results: The differences in the five imaging parameters evaluated by CT, MRI and US between LC and CH were statistically significant (p < 0.001) except for the manifestations of portal hypertension on US. Irregular or nodular surface, blunt edge or morphological changes in the liver were selected as the best predictive signs for cirrhosis on US whereas liver parenchymal abnormalities, manifestations of portal hypertension and morphological changes in the liver were the best predictive signs on MRI and CT by multivariate analysis. The predictive diagnostic accuracy, sensitivity and specificity in discriminating LC from CH based on the best predictive signs were 71.9, 77.1 and 67.6% by CT; 67.9, 67.5 and 68.3% by MRI, and 66.0, 38.4 (lower than CT and MRI, p =0.001) and 88.8% (higher than CT and MRI, p =0.001)by US. According to the imaging impression scoring system, diagnostic accuracy, sensitivity and specificity were 67.0, 84.3 and 52.9% by CT; 70.3, 86.7 and 53.9% by MRI, and 64.0, 52.4 (lower than CT and MRI, p =0.0001) and 73.5% (higher than CT and MRI, p < 0.003) by US. ROC analysis showed that MRI and CT were slightly superior to US in the diagnosis of LC but no statistically significant difference was found between them. For the pathological diagnosis of P-LC, cirrhosis was diagnosed in 59.5, 46.7 and 41.7% of the P-LC cases by US, CT and MRI, respectively, with no significant difference among these methods.

Conclusion: US, CT and MRI had different independent predictive signs for the diagnosis of LC. MRI and CT were slightly superior to US in predicting cirrhosis, especially regarding sensitivity. Noninvasive imaging techniques play an important role in the diagnosis of cirrhosis, especially in the evaluation of P-LC.