Spectral density variation mapping of cerebral waves by three-dimensional interpolation techniques

Abstract

This work focuses on interpolation methods which are proposed as solutions to the EEG source localization. First, a low pass and a high pass filter were applied to the EEG signal in order to remove EEG artifacts. Then, classical interpolation techniques such as three‐dimensional (3D) K‐nearest neighbor and 3D spline were implemented. The major contribution of this article is to develop a new interpolation method called 3D multiquadratic technique which is based on the Euclidean distances between the electrodes. A substitution of the Euclidean distance by the corresponding arc length was realized to promote the 3D spherical multiquadratic interpolation. Based on measured EEG recordings from 19 electrodes mounted on the scalp, these interpolation methods (3D K‐nearest neighbor, 3D spline, 3D multiquadratic and spherical multiquadratic) were applied to EEG recordings of 15 healthy subjects at rest and with closed eyes. The aim of EEG interpolation is to reach the maximum of the spatial resolution of EEG mapping by predicting the brain activity distribution of 109 virtual points located on the scalp surface. The evaluation of the different interpolation methods was achieved by measuring the means of the normalized root mean squared error (NRMSE) and processing time. The results showed that the multiquadratic and 3D spline interpolation methods gave the minimum normalized root mean squared error, but the multiquadratic method was characterized by the minimal processing time compared with 3D K‐nearest neighbor, 3D spline, and 3D spherical multiquadratic methods. Finally, a Spectral density variation mapping of different cerebral waves (delta, theta, alpha and beta) with 128 electrodes was generated by applying the Fast Fourier Transform (FFT). © 2015 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 25, 191–198, 2015