Analysis of risk factors correlated with angiographic vasospasm in patients with aneurysmal subarachnoid hemorrhage using explainable predictive modeling. (September 2021)

• Record Type: Journal Article
• Authors: Kim, Kwang Hyeon; Koo, Hae-Won; Lee, Byung-Jou; Sohn, Moon-Jun
• Is Part Of: Journal of clinical neuroscience. Volume 91(2021)
• Page Start: 334
• Page End: 342
• Publication Date: 2021-09
• Subjects: Aneurysm -- Subarachnoid hemorrhage -- Explainable artificial intelligence -- Feature analysis -- Cerebral arterial vasospasm; Brain -- Surgery -- Periodicals; Neurosciences -- Periodicals; Nervous system -- Surgery -- Periodicals; Brain -- surgery -- Periodicals; Neurosurgical Procedures -- Periodicals; Neurosciences -- Periodicals; Electronic journals; 616.8
• DOI: 10.1016/j.jocn.2021.07.028
• Languages: English
• ISSNs: 0967-5868

Abstract

Graphical abstract: Highlights: This study evaluated the contribution of risk factors correlated with angiographic vasospasm using XAI. Results showed an association between aneurysm size and age as well as angiographic vasospasm. This study provided evidence about risk factors of cerebral vasospasm and the efficacy of machine learning models. Abstract: Cerebral vasospasm (CAV) is a major complication of aneurysmal subarachnoid hemorrhage (aSAH) in patients with ruptured intracranial aneurysm. Explainable artificial intelligence (XAI) was used to analyze the contribution of risk factors on the development of CAV. We obtained data about patients (n = 343) treated for aSAH in our hospital. Predictive factors including age, aneurysm size, Hunt and Hess grade, and modified Fisher grade were used as input to analyze the contribution and correlation of factors correlated with CAV using a random forest regressor. An analysis conducted using an XAI model showed that aneurysm size (27.6%) was most significantly associated with the development of CAV, followed by age (20.7%) and Glasgow coma scale score (7.1%). In some patients with an estimated artificial intelligence-selected CAV value of 51%, the important risk factors were aneurysm size (9.1 mm) and location, and hypertension is also considered a major influencing factor. We could predict that Fisher grade 3 contributed to 20.3%, and the group using Antiplatelet contributed to 12.2% which is expected to lower cerebral CAV compared to the Control group (16.9%). The accuracy rate of the XAI system was 85.5% (area under the curve = 0.88). Using the modeling, aneurysm size and age were quantitatively analyzed and were found to be significantly associated with CAV in patients with aSAH. Hence, XAI modeling techniques can be used to analyze factors correlated with CAV by schematizing prediction results in some patients. Moreover, poor Fisher grade and use of postoperative antiplatelet agent are important factors for prediction of CAV.