Differentiation of intestinal tuberculosis and Crohn's disease through an explainable machine learning method

Area of Science:

• Gastroenterology
• Medical Informatics
• Machine Learning

Background:

• Distinguishing Crohn's disease (CD) from intestinal tuberculosis (ITB) is clinically challenging, impacting treatment strategies.
• Accurate differentiation is essential for appropriate patient management and therapeutic decisions.

Purpose of the Study:

• To develop an explainable machine learning (ML) framework for distinguishing CD from ITB.
• To compare the performance of ML models against traditional statistical methods.
• To provide interpretable insights into the ML model's predictions using the SHAP method.

Main Methods:

• A cohort of 200 patients (160 CD, 40 ITB) was analyzed.
• Feature selection identified nine key variables, including clinical and microbiological markers.
• The XGBoost algorithm was employed and compared with other ML classifiers and statistical methods.

Main Results:

• XGBoost demonstrated superior performance with an Area Under the Curve (AUC) of 0.891.
• The model achieved high sensitivity (0.813), specificity (0.969), precision (0.867), and MCC (0.801).
• SHAP analysis provided effective global and individual patient-level explanations for the XGBoost predictions.

Conclusions:

• An interpretable ML framework, particularly using XGBoost, can effectively differentiate CD from ITB.
• The explainability feature (SHAP) enhances the clinical utility of the ML model.
• This approach offers a valuable tool for improving diagnostic accuracy in challenging gastrointestinal cases.