Improvement of Resting-state EEG Analysis Process with Spectrum Weight-Voting based on LES

TL;DR

This paper introduces a spectrum weight-voting method based on LES for EEG analysis, improving classification accuracy of schizophrenia phases by emphasizing influential frequency bands and aiding psychopathology studies.

Contribution

It proposes a novel weight-voting approach to determine band influence in EEG classification, enhancing accuracy and interpretability in schizophrenia phase detection.

Findings

High correlation between low gamma band weights and schizophrenia phases

Revised features based on weights improve classification accuracy

Band weight distribution aids in understanding schizophrenia risk factors

Abstract

EEG is a non-invasive technique for recording brain bioelectric activity, which has potential applications in various fields such as human-computer interaction and neuroscience. However, there are many difficulties in analyzing EEG data, including its complex composition, low amplitude as well as low signal-to-noise ratio. Some of the existing methods of analysis are based on feature extraction and machine learning to differentiate the phase of schizophrenia that samples belong to. However, medical research requires the use of machine learning not only to give more accurate classification results, but also to give the results that can be applied to pathological studies. The main purpose of this study is to obtain the weight values as the representation of influence of each frequency band on the classification of schizophrenia phases on the basis of a more effective classification method using the LES feature extraction, and then the weight values are processed and applied to improve the accuracy of machine learning classification. We propose a method called weight-voting to obtain the weights of sub-bands features by using results of classification for voting to fit the actual categories of EEG data, and using weights for reclassification. Through this method, we can first obtain the influence of each band in distinguishing three schizophrenia phases, and analyze the effect of band features on the risk of schizophrenia contributing to the study of psychopathology. Our results show that there is a high correlation between the change of weight of low gamma band and the difference between HC, CHR and FES. If the features revised according to weights are used for reclassification, the accuracy of result will be improved compared with the original classifier, which confirms the role of the band weight distribution.