Compressive Sensing-Based Continuous EEG Monitoring: Seizure Detection Performance Comparison of Different Classifiers
Compressive sensing (CS) is a newer sensing modality, which samples the signals at a rate much below the Nyquist rate and still allows the faithful reconstruction from fewer samples. Acquisition of EEG signals using Nyquist sampling generates too many sam
- PDF / 409,739 Bytes
- 10 Pages / 439.37 x 666.142 pts Page_size
- 70 Downloads / 189 Views
Abstract Compressive sensing (CS) is a newer sensing modality, which samples the signals at a rate much below the Nyquist rate and still allows the faithful reconstruction from fewer samples. Acquisition of EEG signals using Nyquist sampling generates too many samples, which invokes the need for compression before storage and transmission of these samples. In this scenario, CS has been proved to be better candidate, suppressing the need of compression by generating the fewer samples, which can be stored or transmitted directly. In this paper, at the transmitter side, the acquisition of EEG signal is done using CS, and then, at the receiver side, the reconstruction is performed using orthogonal matching pursuit (OMP) algorithm of CS. After reconstructing the signal for different undersampling factors, the features are extracted from these signals. Several classifiers are trained and tested on these features to detect the epileptic seizure. Performance comparison of these classifiers shows that even at higher undersampling factors like 64, a high seizure detection accuracy of 98.9% by these classifiers. All the simulations are done on the EEG signal taken from CHB-MIT database using MATLAB 2017a. Keywords Compressive sensing · EEG-monitoring · Random demodulator · OMP · Classifiers
1 Introduction Requirement of continuous EEG monitoring for early detection of disease poses certain challenges on traditional signal acquisition techniques. For faithful reconstruction, sampling such signals by conventional methods results in huge amount M. Rani (B) · S. B. Dhok · R. B. Deshmukh Visvesvaraya National Institute of Technology, Nagpur 440010, MH, India e-mail: [email protected] S. B. Dhok e-mail: [email protected] R. B. Deshmukh e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 D. Harvey et al. (eds.), Advances in VLSI, Communication, and Signal Processing, Lecture Notes in Electrical Engineering 683, https://doi.org/10.1007/978-981-15-6840-4_37
459
460
M. Rani et al.
of samples, and hence, compression is mandatory before storage and transmission. Recently, a rapid growth has been reported in the area of remote health care, in which health is monitored remotely using battery operated devices. Longer battery life is major requirement of these systems for continuous monitoring. But, the difficulty is the requirement of high sampling rate for error-free operation consumes a considerable amount of power, and as a consequence of it, the huge amount samples generated has to be compressed, which requires extra power for further processing. In such situation, compressive sensing (CS) proves itself to be a better candidate to overcome the disadvantages of conventional methods. The acquisition scheme offered by compressive sensing, samples a signal at much lower rate, generating far fewer samples and avoids the need of compression, thereby lowering the power consumption. In literature, use of CS for EEG monitoring has been reported, addressing the applicability of CS for remote heath care [1–7]. Compres
Data Loading...
