An efficient approach for EEG sleep spindles detection based on fractal dimension coupled with time frequency image. (March 2018)

• Record Type: Journal Article
• Authors: Al-salman, Wessam; Li, Yan; Wen, Peng; Diykh, Mohammed
• Is Part Of: Biomedical signal processing and control. Volume 41(2018)
• Page Start: 210
• Page End: 221
• Publication Date: 2018-03
• Subjects: Sleep spindles -- Time frequency image -- Fractal dimension -- Box counting -- EEG signals; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2017.11.019
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Developing a new method to detect sleep spindles based on a fractal technique and a time frequency image. The results showed that FD of time frequency images gives better results in EEG sleep spindles detection than using histograms of TFM. Our findings demonstrated that using TFI based on FD is effective to detect sleep spindles in EEG signals. Abstract: Detection of the characteristics of the sleep stages, such as sleep spindles and K-complexes in EEG signals, is a challenging task in sleep research as visually detecting them requires high skills and efforts from sleep experts. In this paper, we propose a robust method based on time frequency image (TFI) and fractal dimension (FD) to detect sleep spindles in EEG signals. The EEG signals are divided into segments using a sliding window technique. The window size is set to 0.5 s with an overlapping of 0.4 s. A short time Fourier transform (STFT) is applied to obtain a TFI from each EEG segment. Each TFI is converted into an 8-bit binary image. Then, a box counting method is applied to estimate and discover the FDs of EEG signals. Different sets of features are extracted from each TFI after applying a statistical model to the FD of each TFI. The extracted statistical features are fed to a least square support vector machine (LS_SVM) to figure out the best combination of the features. As a result, the proposed method is found to have a high classification rate with the eight features sets. To verify the effectiveness of the proposed method, different classifiers, including a K-means, Naive Bayes and a neural network, are also employed. In this paper, the proposed method is evaluated using two publically available datasets: Dream sleep spindles and Montreal archive of sleep studies. The proposed method is compared with the current existing methods, and the results revealed that the proposed method outperformed the others. An average accuracy of 98.6% and 97.1% is obtained by the proposed method for the two datasets, respectively.