A novel stroke classification model based on EEG feature fusion

Stroke is the leading cause of disability and death worldwide. It severely affects patients' quality of life and imposes a huge burden on the society in general. The diagnosis of stroke relies predominantly on the use of neuroimaging. The identification of stroke using electroencephalogram (EEG) in the clinical assessment of stroke has been underutilized. An EEG feature fusion based light gradient-boosting machine (LightGBM) model was proposed to achieve a fast diagnosis of non-stroke, ischemic stroke, and hemorrhagic stroke. This study aims to capture the essential difference between non-stroke, ischemic stroke, and hemorrhagic stroke. An optimal fusion feature set originated from approximate entropy and fuzzy entropy of EEG signal was constructed. To verify the effectiveness of the EEG fusion feature, the Tree-structured Parzen Estimator optimized LightGBM classifier (TPELGBM) was used for the classification. The ZJU4H EEG dataset used for analysis in this study was obtained from the Fourth Affiliated Hospital of Zhejiang University, China. The proposed ApFu-TPELGBM model exhibited excellent classification results, which achieved a precision of 0.9676, recall of 0.9669, and f1-score of 0.9672. To our knowledge, it was the most accurate classifier for EEG-based stroke diagnosis so far. The ApFu-TPELGBM model can determine the stroke type anywhere EEG signals can be collected, even before the patient is admitted to a hospital. Rapid and accurate diagnosis of stroke using EEG signals may become a promising approach in the clinical assessment of stroke.