Utilize a machine learning approach to predict the development of Delayed Cerebral Ischemia (DCI) in aneurysmal subarachnoid hemorrhage (aSAH) patients using radiological, clinical, and laboratory data.

Delayed Cerebral Ischemia, defined as symptomatic vasospasm or infarction on imaging, is a leading cause of morbidity and mortality in patients with aSAH accounting for 23% of deaths and persistent neurological deficit in 37% of survivors. DCI occurs in about 20-30% of patients with aSAH, and is a sequelae of cerebral vasospasm. DCI may present from 24 hours to 7-10 days post SAH and is difficult to predict. While imaging and clinical features may capture cerebral vasospasm and help predict occurrence, little is known about prediction of DCI. Previous models have not successfully standardized and weighted features through an artificial intelligence algorithm. There is no validated single tool in practice that accounts for the diversity in indicators that may be predictive of DCI. 

Aneurysmal SAH patients who were admitted to the Carle Foundation Hospital from 01/2016 to 12/2021 were retrospectively analyzed. Total of 52 patients met study criteria, out of which 13 (25%) patients had DCI. Relevant data was collected and feature selection and hyperparameter optimization were performed to enhance model efficacy. Models were validated with 10-fold cross validation.

Utilizing Best-First feature selection algorithm, following features were significant at 95% significance level (p<0.05): Transcranial Doppler data, Fisher/Hunt-Hess score, aneurysmal diameter, absolute neutrophils, WBC, BUN, glucose, best motor/verbal response on GCS, diabetes, coronary artery disease and arrhythmia. The Naive Bayes model (ROC-AUC: 0.787) outperformed other algorithms: Logistic Regression (0.753), Multilayer perceptron (0.714), Random Forest (0.751), Logistic Model Tree (0.757). 

An AI-powered framework for early DCI risk assessment created with Naive Bayes ROC AUC of 0.787.