An Explainable Artificial Intelligence Model to Predict Malignant Cerebral Edema after Acute Anterior Circulating Large Hemisphere Infarction

Abstract

Abstract Background: Malignant cerebral edema (MCE) is a serious complication and the main cause of poor prognosis in large hemisphere infarction (LHI). Therefore, rapid and accurate identification of potential patients with MCE is essential for providing timely therapy. However, most prediction models lack interpretability, limiting their use in clinical practice.To establish an interpretable model to predict MCE in patients with LHI. We utilize the SHapley Additive exPlanations (SHAP) method to explain the eXtreme Gradient Boosting (XGBoost) model and identify prognostic factors, providing valuable data for clinical decision-making. Methods: In this retrospective cohort study, we included 314 consecutive patients with LHI admitted to the Third Affiliated Hospital of Soochow University from December 2018 to April 2023. The patients were divided into MCE and non-MCE groups, and we developed an explainable artificial intelligence prediction model. The dataset was randomly divided into two parts: 75% of the data were used for model training and 25% were used for model validation. Confusion matrix was utilized to measure the prediction performance of the XGBoost model. The SHAP method was used to explain the XGBoost model. Decision curve analysis was performed to evaluate the net benefit of the model. Results: A 38.5% (121/314) incidence of MCE was observed among the 314 patients with LHI. The XGBoost model showed excellent predictive performance, with an area under the curve of 0.916 in validation. The SHAP method revealed the top 10 predictive variables of MCE based on their importance ranking, while the Alberta Stroke Program Early CT Score (ASPECTS) score was considered the most important predictive variable, followed by National Institutes of Health Stroke Scale (NIHSS) score, Collateral Status (CS) score, APACHE II score, glycated hemoglobin (HbA1c), atrial fibrillation (AF), neutrophil-to-lymphocyte ratio (NLR), platelet (PLT) count, Glasgow Coma Scale (GCS) and Age. We found that ASPECTS score < 6, NIHSS score >17, CS score < 2, APACHE II >14, HbA1c >6.3 and AF were associated with increased risks of malignant cerebral edema. Conclusion: An interpretable predictive model can increase transparency and help doctors to accurately predict the occurrence of MCE in patients with LHI, providing patients with better treatment strategies and enabling optimal resource allocation.