A New Perspective on Visualising EEG Signal of Post-Stroke Patients

Abstract

Abstract To date, numerous methods have been developed in response to the EEG signal classification of post-stroke patients, among which feature extraction methods are of particular interest. This paper presents a new perspective on the visualisation of the EEG signal of different post-stroke patients in the image representation that can be used to assist in the classification phase. The new perspective for extracting and visualising EEG sub-band features considers the sequential application of power spectral density (PSD) represented in the kernel distribution estimation (KDE) of the PSD manifold. Experiments conducted on 45 post-stroke patients; 14 early, 17 intermediate and 14 advanced patients demonstrated the potential of the proposed perspective to estimate significant parameters under spectral pattern image representation. Visual representation of this new approach shows that the pattern and relationship of post-stroke patients can be clearly visualised. Significant performance can be achieved by classifying post-stroke patients into early-advanced or early intermediate classes as they reach a perfect dissimilarity score, r = 1.00. In the meantime, the absence of beta or theta in pairs has relatively consistent performance in classifying post-stroke patients using sub-bands, and the combination of the two has shown the worst results among other pairs. This paradigm should be included in the future context of the EEG signal classification of post-stroke classes, which could better explain the importance of image representation while improving the accuracy of the specified network.