Machine learning diagnostic framework for liver fibrosis in chronic hepatitis B based on routine laboratory tests

Area of Science:

• Hepatology
• Medical Informatics
• Biostatistics

Background:

• Early detection of liver fibrosis in chronic hepatitis B (CHB) is vital for patient outcomes.
• Accurate fibrosis staging guides treatment decisions and prognosis.
• Current diagnostic methods may have limitations in sensitivity or accessibility.

Purpose of the Study:

• To develop and validate a machine learning (ML)-based model for assessing liver fibrosis degree in CHB patients.
• To compare the performance of ML models against traditional scoring systems.
• To establish a personalized and dynamic risk assessment tool for CHB-associated liver fibrosis.

Main Methods:

• A cohort of 268 CHB patients undergoing liver biopsy was analyzed.
• Liver fibrosis was staged using the Scheuer scoring system.
• Feature selection was performed using least absolute shrinkage and selection operator regression, followed by evaluation of 8 ML algorithms, including random forest (RF).

Main Results:

• The random forest (RF) model demonstrated the best performance, achieving an area under the curve (AUC) of 0.810 (training set) and 0.793 (validation set).
• Decision curve analysis indicated the RF model's superior net benefit across various probability thresholds.
• At a 30% probability threshold, the RF model achieved highest sensitivity and net benefit, outperforming traditional scores.

Conclusions:

• Machine learning models, particularly RF, can effectively assess liver fibrosis in CHB patients.
• The developed RF model offers a safe, effective, and personalized screening method.
• This approach enables earlier and more dynamic risk stratification for CHB patients.