Imaging prediction of nonalcoholic steatohepatitis using computed tomography texture analysis

Area of Science:

• Radiology
• Hepatology
• Medical Imaging Analysis

Background:

• Nonalcoholic steatohepatitis (NASH) is a progressive liver disease.
• Accurate diagnosis of NASH often requires liver biopsy.
• Non-invasive imaging biomarkers are needed for NASH detection.

Purpose of the Study:

• To evaluate the efficacy of non-contrast-enhanced CT (NECT) texture analysis in predicting NASH.
• To develop a predictive model for NASH using CT texture parameters.
• To assess the impact of liver fibrosis on the predictive performance of NECT texture analysis.

Main Methods:

• Retrospective analysis of NECT images from 88 patients with suspected NASH.
• Texture feature extraction with and without filtration.
• Development of a predictive model using a learning dataset and validation on a separate dataset.
• Stratification of patients based on suspected liver fibrosis using hyaluronic acid levels.

Main Results:

• In patients without suspected fibrosis, the NECT texture model achieved an AUC of 0.94 and 94% accuracy.
• Texture parameters 'mean without filter' and 'skewness with 2-mm filter' were key predictors in this group.
• In patients with suspected fibrosis, the model showed lower performance (AUC 0.60, accuracy 42%).

Conclusions:

• NECT texture analysis is an effective tool for predicting NASH in patients without suspected liver fibrosis.
• The presence of hepatic fibrosis may obscure the texture features indicative of NASH.
• This imaging approach offers a potential non-invasive method for NASH assessment.