Deep Learning Radiomics for Liver Fibrosis Staging
Objective
Deep Learning Radiomics study aimed to assess the performance of Deep Learning Radiomics of Elastography (DLRE) for evaluating liver fibrosis stages in patients with chronic hepatitis B. DLRE uses radiomic strategies to analyze the heterogeneity in two-dimensional shear wave elastography (2D-SWE) images, providing a non-invasive method for fibrosis staging.
Design
This prospective multicenter study involved 654 patients from 12 hospitals, with 398 patients and 1990 images ultimately included for analysis. The study compared DLRE’s diagnostic performance to 2D-SWE, aspartate transaminase-to-platelet ratio index, and fibrosis index based on four factors, using liver biopsy as the reference standard. The analysis used receiver operating characteristic (ROC) curves to determine the area under the curve (AUC) for different fibrosis stages, including cirrhosis (F4), advanced fibrosis (≥F3), and significant fibrosis (≥F2). The robustness and accuracy of DLRE were evaluated by varying the number of image acquisitions and training cohorts.
Results
The results showed that DLRE performed exceptionally well across all fibrosis stages. The AUCs for DLRE were:
- 0.97 for cirrhosis (F4)
- 0.98 for advanced fibrosis (≥F3)
- 0.85 for significant fibrosis (≥F2)
These results were significantly superior to other methods, except 2D-SWE in the ≥F2 stage. DLRE’s diagnostic accuracy improved when multiple images (especially three or more) were acquired from each patient. Additionally, the performance remained consistent even with different training cohorts.
Conclusion
DLRE demonstrated superior overall performance in diagnosing liver fibrosis stages compared to 2D-SWE and traditional biomarkers. This technology is promising for the non-invasive, accurate assessment of liver fibrosis in HBV-infected patients, making it a valuable tool in clinical practice for liver disease management.
