Voxel-ICA for reconstruction of source signal time-series and orientation in EEG and MEG
In electroencephalography (EEG) and magnetoencephalography signal processing, scalar beamformers are a popular technique for reconstruction of the time-course of a brain source in a single time-series. A prerequisite for scalar beamformers, however, is that the orientation of the source must be known or estimated, whereas in reality the orientation of a brain source is often not known in advance and current techniques for estimation of brain source orientation are effective only for high signal-to-noise ratio (SNR) brain sources. As a result, vector beamformers are applied which do not need the orientation of the source and reconstruct the source time-course in three orthogonal (x, y, and z) directions. To obtain a single time-course, the vector magnitude of the three orthogonal outputs of the beamformer can be calculated at each time point (often called neural activity index, NAI). The NAI, however, is different from the actual time-course of a source since it contains only positive values. Moreover, in estimating the magnitude of the desired source, the background activity (undesired signals) in the beamformer outputs also become all positive values, which, when added to each other, leads to a drop in the SNR. This becomes a serious problem when the desired source is weak. We propose applying independent component analysis (ICA) to the orthogonal time-courses of a brain voxel, as reconstructed by a vector beamformer, to reconstruct the time-course of a desired source in a single time-series. This approach also provides a good estimation of dipole orientation. Simulated and real EEG data were used to demonstrate the performance of voxel-ICA and were compared with a scalar beamformer and the magnitude time-series of a vector beamformer. This approach is especially helpful when the desired source is weak and the orientation of the source cannot be estimated by other means.
KeywordsBeamformer Electroencephalography Independent component analysis Magnetoencephalography Signal-to-noise ratio Time-course reconstruction
This work was funded by a University of Otago Postgraduate Scholarship. The authors thank Simon Knopp for his suggestions and thoughts on this work.
- 1.Spencer M, Leahy RM, Mosher JC, Lewis PS (1992) Adaptive filters for monitoring localized brain activity from surface potential time series. Proc IEEE Asilomar Conf Signal Syst Comput 26:156–160Google Scholar
- 3.Robinson SE, Vrba J (1998) Functional neuroimaging by synthetic aperture magnetometry (SAM), Proceedings of the 11th International Conference on Biomagnetism pp. 302–305Google Scholar
- 7.Greenblatt RE, Ossadtchi A, Pflieger ME (2005) Local linear estimators for the linear bioelectromagnetic inverse problem. IEEE Trans Biomed Eng 53:3403–3412Google Scholar
- 10.Mohamadi YJ, Poudel G, Innes C, Jones R (2012) Performance of beamformers on EEG source reconstruction. Proc Int Conf IEEE Eng Med Biol Soc 34:2517–2521Google Scholar
- 11.Van Veen BD, Buckley KM (1988) Beamforming: a versatile approach to spatial filtering. IEEE Mag Acoust Speech Signal Process 5:4–24Google Scholar
- 15.Quraan MA (2011) Characterization of brain dynamics using beamformer techniques: advantages and limitations. In: Pang EW (ed) Magnetoencephalography. InTech, Toronto, pp 67–92Google Scholar
- 22.Jung TP, Humphries C, Lee TW, Makeig S, McKeown MJ, Iragui V, Sejnowski TJ (1998) Extended ICA remove artifacts from electroencephalographic recordings. Adv Neural Inf Process Syst 10:894–900Google Scholar
- 28.Jonmohamadi Y, Poudel G, Innes C, Jones R (2013) Electromagnetic tomography via source-space ICA, Proc Int Conf IEEE Eng Med Biol Soc 35:37–40Google Scholar
- 30.Scharf LL, Pezeshki A, Van Veen BD, Cox H, Besson O (2006) Eigenvalue beamforming using a multirank MVDR beamformer and subspace selection. Proceedings of 5th Workshop on Defence Application of Signal Process, Queensland, Australia, 10–14 Dec 2006Google Scholar