Noninvasive assessment of significant liver fibrosis in rabbits by spectral CT parameters and texture analysis

Shi, Qi,Zhang, Minguang

DOI: https://doi.org/10.1007/s11604-023-01423-0

2023-04-19

Japanese Journal of Radiology

Abstract:Noninvasive assessment of significant liver fibrosis in rabbits by spectral CT parameters and texture analysis. Thirty-three rabbits were randomly divided into 27 carbon tetrachloride-induced liver fibrosis group and 6 control group. Spectral CT contrast-enhanced scan was performed in batches, and the liver fibrosis was staged according to the histopathological results. The portal venous phase spectral CT parameters [70 keV CT value, normalized iodine concentration (NIC), spectral HU curve slope (λHU)] were measured, and MaZda texture analysis was performed on 70 keV monochrome images. Three dimensionality reduction methods and four statistical methods in B11 module were used to perform discriminant analysis and calculate misclassified rate (MCR), and ten texture features under the lowest combination of MCR were statistically analyzed. Receiver operating characteristic curve (ROC) was used to calculate the diagnostic performance of spectral parameters and texture features for significant liver fibrosis. Finally, the binary logistic regression was used to further screen independent predictors and establish model. A total of 23 experimental rabbits and 6 control rabbits were included, of which 16 had significant liver fibrosis. Three spectral CT parameters with significant liver fibrosis were significantly lower than those of non-significant liver fibrosis (p 0.5, ranges from 0.764 to 0.875. The logistic regression model showed that Perc.90% and NIC could be used as independent predictors, the overall prediction accuracy of the model was 89.7% and the AUC was 0.976. Spectral CT parameters and texture features have high diagnostic value for predicting significant liver fibrosis in rabbits, and the combination of the two can improve its diagnostic efficiency.

radiology, nuclear medicine & medical imaging