Analysis of EEG-Based Stroke Severity Groups Clustering using K-Means

Abstract

Rehabilitation is the essential key to restore motoric function and brain activity for stroke patients. Electroencephalograph (EEG) has been used widely as an alternative tool to monitor the progress of stroke rehabilitation because EEG represents the motoric function during motion. Determining the stroke severity level is also important during rehabilitation program because it gives information to the clinicians before performing rehabilitation. Stroke severity level will determine which rehabilitation programs the patient should take. Therefore, this study aims to classify stroke severity level by using the EEG features which are the Relative Power Ratio Power Spectral Density (RPR-PSD) and Relative Power Ratio Power Percentage (RPR-PP). The data is collected through the collaboration process with Airlangga University Hospital Surabaya (RSUA). The classes of stroke severity level are defined as severe, moderate, and mild. The EEG frequency sub-bands that were analyzed are Alpha Low (8-9 Hz), Alpha High (9-13 Hz), Beta Low (13-17), and Beta High (17-30 Hz). K-Means clustering method is applied to classify the severity level. From the ANOVA significane value, it shows that all groups of severity level from all sub-bands in this study showed p-value <0.05. This means that each severity group can be classified with its characteristics. From the two features that we analyzed, RPR-PSD showed more suitable condition to differenciate group of severity levels among all EEG frequency sub-bands. Furthermore, Alpha High sub-band showed a better condition to be used as an indicator for monitoring rehabilitation process for stroke patients due to its variance value behaviour. The variance value is changing linearly with the change of severity level compared to other sub-bands.