Radiomic Analysis of Contrast‐Enhanced CT Predicts Recompensation in Hepatitis B‐Related Decompensated Cirrhosis

ABSTRACT Aims Accurately predicting recompensation in patients with decompensated hepatitis B‐related cirrhosis is crucial for guiding treatment strategies. As a reliable model for accurately predicting recompensation is currently lacking, this study aimed to develop a contrast‐enhanced computed tom...

Full description

Saved in:
Bibliographic Details
Main Authors: Qiaofeng Chen, Yuanping Fan, Kaini Wu, Tianpan Cai, Chunyu Lan, Mingju Yu, Yunfeng Fu, Qi Zhu, Jianhao Qiu, Xiaodong Zhou
Format: Article
Language:English
Published: Wiley 2025-03-01
Series:Portal Hypertension & Cirrhosis
Subjects:
Online Access:https://doi.org/10.1002/poh2.70001
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:ABSTRACT Aims Accurately predicting recompensation in patients with decompensated hepatitis B‐related cirrhosis is crucial for guiding treatment strategies. As a reliable model for accurately predicting recompensation is currently lacking, this study aimed to develop a contrast‐enhanced computed tomography (CECT) radiomics model specifically for assessing recompensation outcomes in this patient population. Methods A retrospective cohort of 218 patients with decompensated hepatitis B‐related cirrhosis was included. Patients were randomly divided into training (n = 152) and testing cohorts (n = 66) at a 7:3 ratio by random number generator. Radiomic features (N = 2922) were extracted from arterial phase (AP), venous phase, and delayed phase CECT images. Three machine‐learning algorithms were used to develop radiomic signatures. Independent clinical factors were identified using logistic regression. A combined model integrating radiomic signatures and clinical factors was then constructed. The Shapley Additive Explanation (SHAP) method was employed to visualize and interpret model predictions for individual patients. Model performance was evaluated based on calibration, discrimination, and clinical utility. Results A radiomic signature comprising 10 selected features extracted from AP images using logistic regression showed a significant association with recompensation outcomes (p < 0.001). This signature demonstrated strong discrimination, with an area under the curve (AUC) of 0.929 (95% confidence interval [CI], 0.882–0.968) in the training cohort and 0.853 (95% CI, 0.756–0.960) in the testing cohort. Serum total bilirubin and serum albumin levels were independently associated with recompensation outcomes (p = 0.033 and p < 0.001, respectively). The combined model incorporating both radiomic signatures and clinical factors exhibited comparable predictive performance to the radiomic signature alone, with AUCs of 0.925 (95% CI, 0.871–0.964; p = 0.487) in the training cohort and 0.865 (95% CI, 0.762–0.963; p = 0.669) in the testing cohort. SHAP analysis provided insights into the interpretability of the radiomics model, highlighting the importance of specific radiomic features in predicting recompensation. All prediction models demonstrated good calibration. Decision curve analysis confirmed the clinical utility of the radiomic signature. Conclusions This study demonstrates the strong potential of a CECT‐based computational approach, integrating clinical factors and radiomic features, for predicting recompensation outcomes in patients with hepatitis B‐related decompensated cirrhosis. However, the addition of clinical factors did not provide statistically significant improvement in predictive performance compared to the radiomic signature alone.
ISSN:2770-5838
2770-5846