Radiomics-based classification models for HBV-related cirrhotic patients with covert hepatic encephalopathy

Area of Science:

• Neuroimaging
• Radiology
• Hepatology

Background:

• Covert hepatic encephalopathy (CHE) is linked to precuneus (PC) abnormalities in cirrhotic patients.
• Radiomics analysis offers a potential method to identify imaging biomarkers for CHE.

Purpose of the Study:

• To apply radiomics to identify significant radiomic features in the PC and its subregions for CHE diagnosis.
• To develop and evaluate classification models for CHE using radiomic features and clinical risk factors.

Main Methods:

• 106 HBV-related cirrhotic patients (54 with CHE, 52 without) underwent 3D T1-weighted MRI.
• Radiomic features were extracted from the PC and its subregions, and discriminative features were identified.
• Logistic regression was used to build classification models incorporating radiomic signatures and clinical data.

Main Results:

• The R-C model, combining 4 radiomic features from the right PC, blood ammonia, and Child-Pugh stage, showed the best performance.
• The R-C model achieved an AUC of 0.926, with 1.000 sensitivity, 0.765 specificity, and 0.848 accuracy in the testing set.
• A radiomics nomogram demonstrated good predictive accuracy for CHE in the testing set.

Conclusions:

• Radiomic features of the right PC serve as potential imaging markers for CHE.
• A radiomics nomogram integrating radiomic signatures and clinical factors may aid in individualized CHE prediction.