Analysis on Intraoperative Electrocorticogram Characteristics for Evaluating the Risk of Postoperative Epilepsy*

Epilepsy is one of the most common complications after craniotomy, which happens suddenly and does great harm. There still lacks of effective prediction method during the operation. The main purpose of this paper is to explore the correlation between the characteristics of intraoperative electrocorticogram (ECoG) and postoperative epilepsy, and select effective features to establish a prediction model. This retrospective study uses intraoperative ECoG recordings of 144 patients with cerebrovascular diseases undergoing cerebral revascularization surgeries. The cases are divided into subtypes of ischemic and hemorrhagic. Nine types of ECoG features are designed on different frequency bands indicating clinical information, power spectrum, complexity, sequence change, and information quantity, while their changes in different surgical stages are also considered. Then statistical analysis is used to obtain features significantly related to postoperative epilepsy (p<0.05). The sparse representation method is used on these features to further screen and reduce the redundancy, and then machine learning methods are used to establish a prediction model for postoperative epilepsy. The accuracy, sensitivity and specificity of the best prediction model can achieve 0.817, 0.800 and 0.833 respectively under 5-fold cross validation.Clinical Relevance—This study explores the correlation between the characteristics of intraoperative ECoG and postoperative epilepsy, investigates the possibility to use the ECoG features and machine learning algorithms to assess the risk of postoperative epilepsy during the surgery. Further results are expected to provide reference for preventive measures to reduce the occurrence of postoperative epilepsy.