Independent component analysis (ICA) has been widely used to attenuate interference caused by noise components from the electromagnetic recordings of brain activity. However, the scalp topographies and associated temporal waveforms provided by ICA may be insufficient to distinguish functional components from artifactual ones.
In this work, we proposed two component selection methods, both of which first estimate the cortical distribution of the brain activity for each component, and then determine the functional components based on the parcellation of brain activity mapped onto the cortical surface.
Among all independent components, the first method can identify the dominant components, which have strong activity in the selected dominant brain regions, whereas the second method can identify those inter-regional associating components, which have similar component spectra between a pair of regions. For a targeted region, its component spectrum enumerates the amplitudes of its parceled brain activity across all components. The selected functional components can be remixed to reconstruct the focused electromagnetic signals for further analysis, such as source estimation. Moreover, the inter-regional associating components can be used to estimate the functional brain network.
The accuracy of the cortical activation estimation was evaluated on the data from simulation studies, whereas the usefulness and feasibility of the component selection methods were demonstrated on the magnetoencephalography data recorded from a gender discrimination study.
This technique can ease IC selection by revealing the dominant components corresponding to dominant regions, which contain major brain activity.
Figure 1 : Results of the dominant component selection and the extended dominant component selection method in the gender discrimination study. The temporal waveforms labeled with peak latencies and scalp topographies of dominant components are shown in (a). For IC58, its estimated cortical activation topography and the corresponding scalp projection are illustrated in (c) and (b), respectively. The ACC between the scalp topography a58 and the scalp projection of the estimated cortical activation topography Lb58 is shown in parenthesis. For (d) the right SMG and (e) the right IPL, the remixed signals of their extended dominant components are shown on the left. The scalp topographies at peak latency are shown in the middle and their corresponding source distributions estimated by the MCB method are shown on the right.
Inter-regional associating components
The selection of functional components according to the estimated association matrix may be able to facilitate focusing source estimation on brain activity pertaining to specific functions, such as face processing in the gender discrimination study.
Figure 2 : Source estimation results of (a) the MEG data and (b) the signals remixed from the inter-regional associating components of left IFG (triangular part) and left MTG. The four columns from left to right illustrate the MEG signals, scalp topographies at peak latency, source distribution estimated at peak latency, and the pair of regions involved. The black boxes indicate the 20 ms time windows for estimating brain activation sources using the MCB method.
H.-L. Chan, Y.-S. Chen, L.-F. Chen, T.-H. Chen, and I-T. Chen, ” Lead field space projection for spatiotemporal imaging of independent brain activities ”, 6th ISNN Advances in Neural Networks, 2009.
H.-L. Chan, Y.-S. Chen, and L.-F. Chen, ” Selection of independent components based on cortical mapping of electromagnetic activity ”, J. Neural Eng., Vol. 9, No. 5, Aug. 2012, ID 056006 (18pp). [SCI]