An interpretable machine learning model for diagnostic classification of liver cancer using multivariable clinical
Dianyu Wang1,2, Xiao Li2,3, Zuheng Wang2,4
1Department of Urology, The Second Affiliated Hospital of Fujian Medical University Quanzhou 362000, Fujian, China.
American Journal of Cancer Research
|August 14, 2026
Summary
Machine learning models can aid in liver cancer diagnosis using clinical data. Extreme Gradient Boosting showed high accuracy, with carbohydrate antigen 19-9 being a key predictor for distinguishing liver cancer.
Area of Science:
- Oncology
- Medical Informatics
- Biostatistics
Background:
- Accurate liver cancer diagnosis is crucial for effective treatment.
- Distinguishing liver cancer from other liver diseases using clinical data presents challenges.
Purpose of the Study:
- To develop and validate a machine learning model for liver cancer diagnosis.
- To identify key clinical variables contributing to model performance.
Main Methods:
- Retrospective analysis of 3,629 individuals with suspected liver disease.
- Utilized least absolute shrinkage and selection operator regression for predictor selection.
- Trained and tested nine machine learning algorithms with 10-fold cross-validation and external validation.
Main Results:
- Extreme Gradient Boosting (EGB) demonstrated superior discriminative performance (AUC 1.000 training, 0.937 validation).
- Key predictors included carbohydrate antigen 19-9, alpha-fetoprotein indicators, liver function, and inflammatory markers.
- SHapley Additive exPlanations identified carbohydrate antigen 19-9 as the most significant contributor.
Conclusions:
- Machine learning, especially EGB, can integrate clinical indicators for liver cancer diagnosis.
- The developed model shows potential as an auxiliary tool for clinical assessment.
- Prospective validation is needed for early risk prediction and screening applications.