Developing an interpretable machine learning model for diagnosing gout using clinical and ultrasound features.

Area of Science:

• Rheumatology
• Artificial Intelligence
• Medical Imaging

Background:

• Gout diagnosis relies on clinical assessment and imaging, but early prediction remains challenging.
• Machine learning offers potential for improving diagnostic accuracy by integrating diverse data types.

Purpose of the Study:

• To develop and interpret a machine learning (ML) model for gout prediction using clinical data and ultrasound features.
• To enhance clinical decision support systems for gout diagnosis.

Main Methods:

• Analysis of first metatarsophalangeal (MTP1) joint ultrasound data from 609 patients across two institutions.
• Development of six ML models, with predictor selection using Random Forest, LASSO, and XGBoost.
• Application of SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• Identification of five key predictors: serum uric acid (SUA), deep learning (DL) model predictions, tophus, bone erosion, and double contour sign (DCs).
• Optimal performance by a logistic regression (LR) model with Area Under the Curve (AUC) of 0.870 (internal) and 0.854 (external).
• The model demonstrated good calibration in both internal and external testing cohorts.

Conclusions:

• An interpretable ML model for gout prediction was successfully developed.
• SHAP analysis elucidated the contribution of key features, supporting clinical decision-making.
• This work provides a foundation for advanced clinical decision support in gout diagnosis.