Machine Learning Predicts the Cerebral Hyperperfusion Syndrome after Combined Bypass Surgery in Adult Moyamoya Disease with High Performance. (16th November 2020)

• Record Type: Journal Article
• Authors: Yamamoto, Shusuke; Hori, Emiko; Kashiwazaki, Daina; Akioka, Naoki; Shibata, Takashi; Akai, Takuya; Kuwayama, Naoya; Kuroda, Satoshi
• Is Part Of: Neurosurgery. Volume 67(2010)Supplement 1
• Publication Date: 2020-11-16
• Subjects: Nervous system -- Surgery -- Periodicals; 617.48005
• DOI: 10.1093/neuros/nyaa447_249
• Languages: English
• ISSNs: 0148-396X

Abstract

Abstract: INTRODUCTION: Although surgical revascularization including direct bypass surgery is the most common and successful therapy to improve cerebral hemodynamics and reduce the risk of subsequent ischemic and hemorrhagic stroke in moyamoya disease (MMD), it sometimes leads to postoperative cerebral hyperperfusion syndrome (CHS). The cause of CHS is still unclear and it is quite difficult to predict the occurrence accurately, probably because several complicated factors are associated with the occurrence. On the other hand, machine learning (ML) is quite useful to predict the multifactorial events in various fields including medicine. METHODS: This retrospective study included 73 involved hemispheres of 51 adult patients with MMD that underwent combined bypass surgery. CHS was diagnosed using single photon emission computed tomography (SPECT) at postoperative day-0, 2, and 7. Important features for creating prediction model were selected from demographic and radiological data based on the importance calculated using ensemble classifier and statistical analysis. Prediction model was created using leave-one-out cross-validation for totally 11 ML algorithms. The performance of prediction models was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and precision. RESULTS: Eleven of 73 hemispheres developed postoperative CHS. The prediction model with the best performance was created using 5 important features including childhood onset, ischemic onset, cerebral blood flow (CBF) impairment in middle cerebral artery (MCA) area, CBF impairment in posterior cerebral artery (PCA) area, and cerebrovascular reactivity (CVR) impairment in PCA area and Gaussian support vector machine (SVM) algorithm demonstrated quite high performance (AUC 0.95, sensitivity 1.00, specificity 0.92, and precision 0.80). CONCLUSION: ML can predict the occurrence of postoperative CHS in MMD with quite high performance. In the best prediction model, childhood onset, ischemic onset, CBF impairment in MCA area, CBF impairment in PCA area, and CVR impairment in PCA area were used as the important features and Gaussian SVM was applied as the best ML algorithm. ML is quite suitable for predicting multifactorial and complicated clinical event.