Prediction of postoperative haemorrhage after cerebral tumour surgery using machine learning algorithms

Background Postoperative intracranial hemorrhage is a critical complication following cerebral tumor surgery, often associated with increased morbidity and mortality. This study aimed to predict the risk of postoperative intracerebral hemorrhage in patients undergoing intracranial tumor surgery by employing machine learning (ML) algorithms for risk stratification and identifying key contributing factors. Methods A retrospective analysis was conducted on 118 patients who underwent intracranial tumor surgery and were monitored in the neurosurgical intensive care unit between January 2024 and January 2025. Patients with radiologically confirmed hematomas ≥ 5 cm³ on brain CT within 24–48 hours postoperatively were classified as "Positive" for bleeding, while others were labeled "Negative." Clinical and biochemical parameters were analyzed using SPSS and R. Multiple ML algorithms—including Bagging MARS, Boosting C5.0, SVM, and Random Forests—were developed and evaluated using performance metrics such as AUC, F-score, accuracy, and Brier score. Results The Bagging MARS model demonstrated superior predictive performance, with a test AUC of 0.8693, accuracy of 80.8%, Brier score of 0.1580, and F-score of 0.8649. Platelet count, serum sodium level, and glomerular filtration rate (GFR) emerged as the most influential predictors of hemorrhage. Model explainability was enhanced using SHAP and LIME analyses, offering both global and local interpretability of the predictions. Conclusion ML algorithms, particularly Bagging MARS, show high accuracy in predicting postoperative hemorrhage following brain tumor surgery. Biomarkers such as platelet count, sodium, and GFR offer clinically meaningful insights for early risk detection and intervention. Integration of these predictive models into clinical decision support systems may significantly improve postoperative monitoring and patient outcomes.