# Enhancing the Diagnosis of Behçet’s Disease Using Machine Learning: A Comparative Study on Clinical Data From Saudi Arabia

## TL;DR

This study uses machine learning to explore features that distinguish Behçet’s disease from other rheumatic conditions using Saudi clinical data.

## Contribution

The study introduces a novel exploratory framework using machine learning to identify key features of Behçet’s disease.

## Key findings

- Random Forest achieved 96.7% accuracy in distinguishing Behçet’s disease from other rheumatic conditions.
- Oral ulcers were identified as the most important feature for diagnosis.
- Machine learning models showed high performance, with XGBoost achieving an AUC of 0.9985.

## Abstract

Behçet’s disease (BD) is one of the most difficult diseases to diagnose in the field of rheumatic immune diseases because it is rare, has many different symptoms, and we do not know much about how it works. Instead of trying to make a direct clinical diagnosis, this study was set up as an exploratory investigation to find out more about BD and figure out which clinical and laboratory features are most important. To accomplish this, clinical data were gathered from 148 patients (76 with bipolar disorder and 72 with rheumatoid arthritis) at the Rheumatology Clinic of King Abdulaziz University. We used several machine learning (ML) algorithms, such as decision tree, bagging, random forest (RF), XGBoost, and support vector machines (SVMs), to see if they could learn patterns that set BD apart from other rheumatic diseases. We used three different methods to find out how important each feature was: built‐in model importance, permutation‐based analysis, and Shapley additive explanation (SHAP) values. The ML models worked well, with the RF getting the best accuracy (96.7%) and an area under the curve (AUC) of 1.0. XGBoost came in second with an AUC of 0.9985. The feature analysis showed that the results were partially in line with established diagnostic criteria (Japan, ISG, and ICBD), with oral ulcers being the most important feature. Overall, this study serves as an exploratory framework to deepen understanding of BD’s distinctive characteristics and underlying feature interactions, offering insights that can inform future diagnostic support systems rather than serving as a diagnostic tool itself.