Related Experiment Video
Updated: Jul 14, 2026
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
Predicting Post-Craniotomy Intracranial Infection Using Perioperative and Early Postoperative Variables: A
Zian Zhong1, Liangjin Zhang1, Ping Yu1
1Department of Neurology, Xiangyang Central Hospital, Affiliated Hospital of Hubei University of Arts and Science, Xiangyang, Hubei, People's Republic of China.
Objective:
To identify perioperative clinical and surgical variables associated with intracranial infection (ICI) after craniotomy, and establish an accurate machine learning prediction model for clinical risk stratification and early intervention.
Methods:
A retrospective study included 419 patients who underwent craniotomy surgery at Xiangyang Central Hospital from May 2017 to April 2024. Perioperative variables were collected. Univariate and multivariate logistic regression were used to screen for risk factors. Four models (LR, RF, XGBoost, LightGBM) were constructed and validated through AUC, sensitivity, specificity, PPV, NPV, calibration curve, and DCA were used to evaluate the performance.
Results:
53 patients (12.6%) developed ICI. Independent risk factors included prolonged surgical time, significant intraoperative bleeding, postoperative cerebrospinal fluid leakage, multiple surgeries (≥ 2), and frequent cerebrospinal fluid sampling/drug injection through drainage tubes. LightGBM achieved the best performance: training set AUC 0.873 (95% CI 0.819-0.927), test set AUC 0.811 (95% CI 0.758-0.924), sensitivity 81.0%, specificity 79.3%, PPV 38.6%, NPV 95.7%. Good calibration and high clinical practicality were confirmed.
Conclusion:
Machine learning models, especially LightGBM, can effectively predict postoperative ICI. The identified risk factors and best models support personalized risk assessment and clinical prevention.
