Insights from machine learning: modeling the clinical care of hemorrhagic brain pawns

In recent years, with the increasing incidence of stroke, hemorrhagic stroke, as a serious neurological emergency, prompt and accurate diagnosis and treatment can significantly reduce mortality and disability rates among patients. Clinical decision support plays a crucial role in the research of hemorrhagic stroke. It integrates radiological features, clinical patient information, treatment protocols, and clinical outcomes to achieve precise and personalized efficacy assessment and prognosis prediction. Assisting clinicians in rapid diagnosis and predicting disease progression allows for timely intervention strategies. Predicting hematoma expansion (HE) and selecting appropriate clinical treatment plans are pivotal in managing hemorrhagic stroke. This study processed data extensively to derive model data directly applicable to exploring risk factors associated with HE. The modified Rankin Scale (mRS) was used to assess the prognosis of hemorrhagic stroke patients. Five machine learning models— Random Forest, XGBoost, Logistic Regression, LGBoost, and AdaBoost—were employed to predict the probability of HE and the prognosis of hemorrhagic stroke patients. Models were further optimized using grid search and ten-fold cross-validation to obtain more accurate predictions. The study identified close correlations among various factors, including mRS scores, and histories of diabetes, coronary heart disease, hematoma volume, and age. RF and XGBoost demonstrated excellent performance in predicting HE probability, with ROC values of 0.98. LGBoost outperformed other models in predicting the prognosis of hemorrhagic stroke patients.