An Explainable Ensemble Learning -Based Auxiliary Diagnosis System for Cerebral Small Vessel Disease

Cerebral small vessel disease (CSVD) poses major public health challenges, yet current MRI based diagnosis detects only established damage, and existing auxiliary methods, mostly based on conventional statistics, lack sufficient feature extraction capability and generalizability, thereby limiting early warning and precision management. Accordingly, we developed an intelligent auxiliary diagnostic system grounded in an interpretable ensemble learning framework, aiming to enable early detection and warning of CSVD. To support this development, a total of 597 sets of electronic medical record data from Quzhou Affiliated Hospital of Wenzhou Medical University were used as the study cohort. Firstly, a multidimensional feature evaluation and selection method was proposed, identifying 12 key predictive factors out of 23 relevant variables. Subsequently, the optimal algorithm was selected from Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, Gradient Boosting Machine, XGBoost, and Multilayer Perceptron Classifier, based on Area Under the Curve (AUC) and Accuracy metrics, and a stacking ensemble learning strategy was then employed for model construction. The developed model demonstrated excellent discriminative performance, achieving an AUC of 0.924 while maintaining a low Brier score of 0.1065. By integrating the SHAP interpretability algorithm, the model provided intuitive visualizations of feature importance, thereby enhancing transparency and facilitating clinical adoption. Ultimately, this study achieved effective integration of early warning and auxiliary diagnostic functions for CSVD. These results indicate that the proposed system possesses high accuracy, interpretability, and deployability, underscoring its broad potential for early warning and personalized management of CSVD.