期刊名称：Lecture Notes in Engineering and Computer Science
印刷版ISSN：2078-0958
电子版ISSN：2078-0966
出版年度：2019
卷号：2240
页码：463-467
出版社：Newswood and International Association of Engineers
摘要：—In this study, EEG signals were converted using
continuous Wavelet transform (CWT) and short-term Fourier
transform (STFT) into time-frequency images as input to the
convolutional neural network. According to Bi-Spectral (BIS)
index and signal quality indicator (SQI) of commercial machines,
anesthetic state can be classified as anesthetic light (AL),
anesthetic ok (AO), anesthetic deep (AD), and Noise. The EEG
signal is converted into an image every 5 seconds as well as 2
minutes period. The 5 seconds images dataset was generated
from 13 patients as reported in a previous study which is
compared to current study that is based on 2 minutes images
dataset generated 55 patients. As a result, the 5 seconds EEG
CWT image model predicts an accuracy of the individual
categories of: AL is 69%, AO is 75%, AD is 73%, and Noise is
50%. The overall accuracy of the model is 72.13%. However, the
2 minutes EEG CWT images model predicts an accuracy of the
individual categories of: AL is 81%, AO is 86%, AD is 91%, and
Noise is 59%. The overall accuracy of the model is 85.62%. In
addition, the 2 minutes EEG STFT image model predicts the
accuracy of individual categories of AL is 82%, AO is 85%, AD
is 92%, and Noise is 52%. The overall accuracy of the model is
84.71%. The result shows that the 2 minutes images model is
better than the 5 seconds images model. Therefore, ten patients
were randomly selected from the data of 55 patients as test data.
The test results show an overall accuracy of 92.5% and 87.85%
for the CWT image model and the STFT image model. In
conclusion, the 2 minutes EEG CWT image model is the best
model for this study.
