Noninvasive Assessment of Liver Fibrosis and Inflammation in Chronic Hepatitis B: A Dual-task Convolutional Neural Network (DtCNN) Model Based on Ultrasound Shear Wave Elastography

BACKGROUND AND AIMS: Liver stiffness (LS) measured by shear wave elastography (SWE) is often influenced by hepatic inflammation. The aim was to develop a dual-task convolutional neural network (DtCNN) model for the simultaneous staging of liver fibrosis and inflammation activity using 2D-SWE. METHODS: A total of 532 patients with chronic hepatitis B (CHB) were included to develop and validate the DtCNN model. An additional 180 consecutive patients between December 2019 and April 2021 were prospectively included for further validation. All patients underwent 2D-SWE examination and serum biomarker assessment. A DtCNN model containing two pathways for the staging of fibrosis and inflammation was used to improve the classification of significant fibrosis (≥F2), advanced fibrosis (≥F3) as well as cirrhosis (F4). RESULTS: Both fibrosis and inflammation affected LS measurements by 2D-SWE. The proposed DtCNN performed the best among all the classification models for fibrosis stage [significant fibrosis AUC=0.89 (95% CI: 0.87–0.92), advanced fibrosis AUC=0.87 (95% CI: 0.84–0.90), liver cirrhosis AUC=0.85 (95% CI: 0.81–0.89)]. The DtCNN-based prediction of inflammation activity achieved AUCs of 0.82 (95% CI: 0.78–0.86) for grade ≥A1, 0.88 (95% CI: 0.85–0.90) grade ≥A2 and 0.78 (95% CI: 0.75–0.81) for grade ≥A3, which were significantly higher than the AUCs of the single-task groups. Similar findings were observed in the prospective study. CONCLUSIONS: The proposed DtCNN improved diagnostic performance compared with existing fibrosis staging models by including inflammation in the model, which supports its potential clinical application.