Background and Objective:Accurate classification of liver disease stages provides crucial insights into patient prognosis, aiding in the prediction of disease outcomes and influencing clinical decision-making. There is an urgent need for non-invasive methods to diagnose various stages of liver dysfunction and uncover hidden pattern based on individual disease characteristics. Method:One popular and effective approach is collecting serum biomarker samples. The study was conducted on collected serum biomaker samples of 81 patients with Inflammatory Bowel Disease (IBD) of Changhua Christian Hospital in China, including 36 with Crohn’s disease (CD) and 45 with Ulcerative Colitis (UC) using Latent Semantic Analysis(LSA) and machine learning (ML) techniques.Machine Learning algorithms Random Forest (RF), Logistic Regression (LR), XGBoost (XGB), and Support Vector Classifier (SVC), were utilized to predict liver risk associated with conditions including Hepatitis, Autoimmune Hepatitis (AIH), Alcoholic Liver Disease (ALD), and Non-Alcoholic Fatty Liver Disease (NAFLD). Models’ accuracy was assessed using K-Fold Cross-Validation (CV).Distinct pattern were identified using Latent Semantic Analysis(LSA). Furthermore, SHAP plots were utilized for enhanced interpretability, highlighting essential features for liver dysfunction levels. Results:The inflammatory profile, mixed disease profile, and healthy profile were the three distinct clusters were identified with LSA. The RF model achieved high accuracy of 0.94±0.06. Serum Glutamate Pyruvate Transaminase (GPT), Age at Diagnosis (AAD), Erythrocyte Sedimentation Rate (ESR), C-reactive protein (CRP) were found the most key important features in liver disease staging increment. Conclusion:The research significantly contributes to the fields of biomedical informatics and clinical decision-making. The developed model offers valuable decision-making tools for clinicians, enabling early and targeted interventions.
Read full abstract