Non-contrast CT radiology-clinical machine learning modeling to predict chronic hydrocephalus after aneurysmal subarachnoid hemorrhage

Objective To establish a machine learning model based on radiomics of non-contrast CT and clinical features to predict the occurrence of chronic hydrocephalus after aneurysmal subarachnoid hemorrhage. Methods A retrospective analysis of 150 patients with aneurysmal subarachnoid hemorrhage (aSAH) who underwent surgery between January 2020 and February 2024 was performed. Chronic hydrocephalus(CHC), defined as hydrocephalus occurring 14 days after ruptured aneurysmal hemorrhage, was determined primarily from follow-up CT images. Radiological features were extracted from non-contrast CT (NCCT) and screened using the least absolute shrinkage and selection algorithm (LASSO) regression method. The logistic regression (LR) model was employed to construct models by leveraging radiomic as well as clinical characteristics. A radiological-clinical nomogram model was developed and the predictive performance of the model was assessed using area under the curve (AUC), accuracy, sensitivity and specificity. Results A total of 150 patients were enrolled in this study. From non-contrast CT scans, 1,834 radiomic features were extracted, with 12 optimal features selected to construct the radiomic model. Univariate and stepwise multivariate analyses identified the Glasgow Coma Scale (GCS) score at admission and posterior circulation aneurysms as independent factors for constructing the clinical model. The radiomic-clinical nomogram model demonstrated area under the curve (AUC) values of 0.860 (95% CI: 0.7906–0.9303) in the training cohort and 0.683 (95% CI: 0.4795–0.8856) in the testing cohort. Conclusion The radiology - clinical nomogram model based on non - contrast CT shows a rather good performance in predicting chronic hydrocephalus following aneurysmal subarachnoid hemorrhage.