Related Experiment Video
Updated: May 21, 2026
Autologous Blood Injection to Model Spontaneous Intracerebral Hemorrhage in Mice
Published on: August 24, 2011
Development and internal validation of a machine learning model for predicting intracranial infection after
Yizhao Lin1, Wentong Zheng1, Dankui Zhang2
1Department of Laboratory Medicine, Dehua County Hospital, Quanzhou, China.
Background:
Intracranial infection (ICI) is a serious complication following spontaneous intracerebral hemorrhage (ICH) and is associated with prolonged intensive care, increased morbidity, and poor functional outcomes. Early identification of patients at high risk for post-ICH ICI remains difficult because of heterogeneous clinical presentations and complex interactions among neurological severity, systemic inflammation, and treatment-related factors. This study aimed to develop and validate a clinically applicable machine learning model for early prediction of ICI after ICH.
Methods:
This two-center retrospective study included 1,317 patients with spontaneous ICH admitted to two hospitals in the same province, between 2015 and 2024. Baseline demographic, clinical, laboratory, and radiological variables obtained within 24 h of admission were used to construct the prediction models. Twelve machine learning algorithms were compared, and a Light Gradient Boosting Machine (LGBM) model demonstrated the best overall performance. Model discrimination, calibration, and clinical utility were evaluated using receiver operating characteristic analysis, calibration plots, precision-recall curves, decision curve analysis, and 10-fold cross-validation. Associations between model-predicted risk, ICI occurrence, and 180-day functional outcomes were assessed.
Results:
Intracranial infection occurred in 165 patients (12.5%). The LGBM model showed excellent test-set discrimination (AUC = 0.923), and supplementary 10-fold cross-validation on the overall cohort suggested relatively stable performance across folds (mean AUC = 0.933). Higher model-predicted risk was independently and nonlinearly associated with increased ICI risk and was significantly associated with unfavorable 180-day functional outcomes.
Conclusion:
This ML model showed good performance for the early prediction of ICI after ICH using routinely available clinical data and may support risk stratification in neurocritical care settings. However, because only internal validation was performed, further external validation is needed before broader clinical application.
