Hybrid Significant Stroke Feature: A Novel Stroke Feature Analysis Approach for Stroke Severity Classification of EEG Signals Based on Time Domain, Frequency Domain, and Signal Decomposition Domain

The assessment of stroke severity plays a crucial role in determining the appropriate medical interventions for stroke patients. It helps identify the necessary steps to be taken during these patients’ monitoring, evaluation, and rehabilitation process. Neurologists solely depend on visualization to assess the extent of a stroke in patients during the examination session. In the process of observing through visualization, significant information is sometimes overlooked. Therefore, it is necessary to supplement visualization with the EEG signal analysis method to detect strokes. This approach presents difficulties because of the signal's non-stationary, non-linear, multi-band, multi-channel characteristics and the separate processing and feature extraction in each domain, leading to a high-dimensional space. This study addresses the challenges of signal processing and feature extraction by proposing a feature analysis methodology. The approach employs a multi-domain hybrid significant stroke feature set and the Hybrid Significant Stroke Feature (HSSF) method to determine stroke severity accurately. The HSSF method employs sampling of alpha low and alpha high sub-bands in the alpha band and beta low and beta high in the beta band. This approach utilizes ten characteristics in the time domain, frequency domain, signal decomposition domain, and three types of motion stimuli, specifically shoulder, elbow, and grab. The signal processing and feature analysis strategy comprises the steps of data collection, data preparation, feature extraction, and selection of salient characteristics for categorization. Dimensionality reduction involves prioritizing the selection of low-dimensional domains. Accuracy enhancement is attained by doing normalcy testing and significant feature testing to determine the most significant mean value. The experimental results demonstrate that the HSSF approach, which incorporates classification based on type, domain, feature, stimulus, and subband, achieves a 98% accuracy in identifying stroke severity through SVM classification of PSD, Hjorth Parameter Mobility, and Hjorth Parameter Complexity features with the high beta sub-band. This data validates that the HSSF technique proficiently detects the optimal parameters within a restricted dimensional space to assess stroke severity precisely.