Beamforming methods [TUTORIAL UNDER DEVELOPMENT: NOT READY FOR PUBLIC USE] Authors: Hui-Ling Chan, Francois Tadel, Sylvain Baillet The estimation of source distribution is an important step to understand the brain activity from EEG and MEG data. Dipole fitting, minimum norm estimation and beamformer are three commonly used methods. It has been proved that beamforming methods provide good spatial resolution. This tutorial will show how to apply beamforming methods to MEG data and obtain the statistic map of source activation. We are going to use the protocol TutorialRaw created in the introduction tutorials. If you have not followed these tutorials yet, please do it now.

Introduction Beamforming methods scan each targeted source position

latex error! exitcode was 2 (signal 0), transscript follows:

and estimate the spatial filter

latex error! exitcode was 2 (signal 0), transscript follows:

. By multiplying with the MEG recordings

latex error! exitcode was 2 (signal 0), transscript follows:

, the spatial filter

latex error! exitcode was 2 (signal 0), transscript follows:

outputs the temporal waveform

latex error! exitcode was 2 (signal 0), transscript follows:

of the dipole source at that position with the dipole orientation

latex error! exitcode was 2 (signal 0), transscript follows:

as below:

latex error! exitcode was 2 (signal 0), transscript follows:

where 'T' indicates the transpose of a matrix or vector. The beamforming spatial filter can be vector-type or scalar-type. Vector-type beamformer For each position

latex error! exitcode was 2 (signal 0), transscript follows:

, three orthogonal spatial filters

latex error! exitcode was 2 (signal 0), transscript follows:

are computed by applying the unit-gain constraint as well as the minimum norm and minimum variance criteria as below [Liu and Van Veen, 1992; Van Veen et al., 1997]:

latex error! exitcode was 2 (signal 0), transscript follows:

where

latex error! exitcode was 2 (signal 0), transscript follows:

is the covariance matrix of MEG recordings during window

latex error! exitcode was 2 (signal 0), transscript follows:

,

latex error! exitcode was 2 (signal 0), transscript follows:

is the identity matrix,

latex error! exitcode was 2 (signal 0), transscript follows:

is the gain matrix for the dipole located at position

latex error! exitcode was 2 (signal 0), transscript follows:

, and

latex error! exitcode was 2 (signal 0), transscript follows:

is the regularization parameter which compromises the minimum norm and minimum variance criteria. Scalar-type beamformer For each position

latex error! exitcode was 2 (signal 0), transscript follows:

, the source orientation

latex error! exitcode was 2 (signal 0), transscript follows:

is first estimated to enable the spatial filter to output the source activity with maximum power or fitting some other criteria. The dipole orientation can be obtained by exhaustive search, non-linear search [Robinson and Vrba, 1999], or analytical solution [Chen et al., 2006]. The estimated dipole orientation

latex error! exitcode was 2 (signal 0), transscript follows:

is then applied to calculate the spatial filter as follows [Robinson and Vrba, 1999]:

latex error! exitcode was 2 (signal 0), transscript follows:

LCMV beamformer LCMV stands for linearly-constrained minimum variance. LCMV beamformer is vector-type beamformer [Van Veen et al., 1997]. For each position

latex error! exitcode was 2 (signal 0), transscript follows:

, this method calculates the neural activity index, which is interpreted as the estimate of source to noise variance. Neural activity index For each position

latex error! exitcode was 2 (signal 0), transscript follows:

, the variance of source activity during active state

latex error! exitcode was 2 (signal 0), transscript follows:

is calculated as follows [Liu and Van Veen, 1992; Van Veen et al., 1997]:

latex error! exitcode was 2 (signal 0), transscript follows:

or

latex error! exitcode was 2 (signal 0), transscript follows:

where

latex error! exitcode was 2 (signal 0), transscript follows:

is the covariance matrix computed from MEG recordings during active state

latex error! exitcode was 2 (signal 0), transscript follows:

. Alternatively, the variance of beamformer outpout along the dominant direction can be calculated by using the singular value decomposition [Gross et al., 2001]:

latex error! exitcode was 2 (signal 0), transscript follows:

or

latex error! exitcode was 2 (signal 0), transscript follows:

where

latex error! exitcode was 2 (signal 0), transscript follows:

indicates the maximum eigenvalue of the expression in braces. When the location

latex error! exitcode was 2 (signal 0), transscript follows:

is far from sensors, the elements of lead field matrix

latex error! exitcode was 2 (signal 0), transscript follows:

are small. So the elements of

latex error! exitcode was 2 (signal 0), transscript follows:

are generally large and the estimated variance for the deep source becomes large. When the location

latex error! exitcode was 2 (signal 0), transscript follows:

is close to sensors, the elements of lead field matrix

latex error! exitcode was 2 (signal 0), transscript follows:

are large. It results in small values of the elements of

latex error! exitcode was 2 (signal 0), transscript follows:

and small estimated variance for the superficial source. To reduce the effect caused by the depth of source location, the estimated source variance

latex error! exitcode was 2 (signal 0), transscript follows:

is normalized by the noise variance

latex error! exitcode was 2 (signal 0), transscript follows:

as follows:

latex error! exitcode was 2 (signal 0), transscript follows:

or

latex error! exitcode was 2 (signal 0), transscript follows:

where

latex error! exitcode was 2 (signal 0), transscript follows:

is the covariance matrix computed from MEG recordings during control state

latex error! exitcode was 2 (signal 0), transscript follows:

. The normalized variance

latex error! exitcode was 2 (signal 0), transscript follows:

is called neural activity index (NAI). In the Brainstorm, the noise covariance will be used as the covariance matrix for control state. Please refer to Tutorial 7: Noise covariance matrix for the details of noise covariance computation. LCMV process

• Drag and drop all of the epochs in Subject01 / right / right | timeoffset (98 files) into Process1.

  Click on [RUN] and select the process "Sources > Beamformer: LCMV".
  Configure the options as follows:
  
  LCMV_options.png With this window you can select the method you want to use to estimate the source variance and noise variance, the time window of active state, the regularization parameter, and the sensors you are going to use for this estimation. You can edit the following options:

  • Comment: This field contains what is going to be displayed in the database explorer. Method:
    • Unconstrained (max singular value): Calculates the variances of source activity and noise by using the maximum singular value. Unconstrained (trace): Calculates the variances by using the trace. Constrained: The dipole orientation for each vertex is set to the normal vector to the cortex surface (not available for volume head models). This method may give less smooth results than the unconstrained methods.

    Time window of active state: The neural activity index is computed using the source activity variance during the active state

    latex error! exitcode was 2 (signal 0), transscript follows:
    
    

    .
    • Time range of interest: Total time period over which we want to calculate the neural activity index. Active window size: Defines the duration of the active state. The active window is slided all along the time range of interest. A large active window size is recommended, at least larger than the number of sensors in number of time samples. Temporal resolution: Defines the time lapse between two estimations of the neural activity index. If this "active size window" or "temporal resolution" are set to zero, the active window size is set to the entire time range of interest. The temporal resolution becomes irrelevant because only one neural activity index will be estimated.

      If there are multiple active state windows, the time label of the neural activity index will be set as the middle time of each active state window, that is,

      latex error! exitcode was 2 (signal 0), transscript follows:
      
      

      in the following figure. You are going to obtain the spatiotemporal dynamics of neural activity index bewteen

      latex error! exitcode was 2 (signal 0), transscript follows:
      
      

      and

      latex error! exitcode was 2 (signal 0), transscript follows:
      
      

      .
      
      slidingwindow.png
    Regularization parameter: This value will be multiplied with the maximum eigenvalue of the covariance matrix computed from the recordings during Time range of interest. A larger value of regularization parameter would give smoother results. Large regularization parameter value is more suitable for noisy data. When analyzing simulation data sets, the regularization parameter value can be reduced. Sensors type: Modalities that are used for the reconstruction. Here we only have one type of MEG sensors, so nothing to change. Save spatial filters: Check this option if you want to save the spatial filters. Checking this option if you want to observe the temporal profile of source or compute functional connectivity. Note that the amplitudes of these temporal profiles are not meaningful because the spatial filter is caluclated for one voxel at a time without the holistic consideration on whole brain activity. So, the amplitude of beamformer outpout can not represent the source amplitude and the relationship of the amplitude between each grid point is meaningless. However, the beamformer outpouts can further be used to calculate correlation or coherence between each region because these two estimations include the normalization of amplitude and thus the ampltidue of source waveforms doesn't effect the results.

  Two new files are available in the database explorer.
  

  • right: LCMV (Unconstr/svd, 0-296ms): This file saves the neural activity index for every dipole position.
    
    LCMV_results_list_svd_source.png
    
    Double click on this file. Right click on anywhere of the pop-out window. From the pop-out menu, select Colormap:Sources > Maximum: Custom.... Set Minimum to be 1.2 and Maximum to be 2.0. Click Ok.
    
    Colormaplimit.png
    
    Set the time to be anytime between 0 and 295.8 ms. Then click Surface tab on the right panel of database explorer. Set Data options > Amplitude to be 50%.
    
    LCMV_results_svdtrace_map.png
    
    The maps of neural activity index estimated using Unconstrained (max sigular value) and Unconstrained (trace) are shown in the left and right figures, respectively. Both maps display strong activations in the left sensory area. right: LCMV/filter (Unconstr/svd,0-296ms): This file saves the three spatial filters along the x, y, and z directions for each dipole position. These spatial filters are normalized by the standard deviation of noise. Note that there is one spatial filter for each dipole location if the method Cortically Constrained is used.
    
    LCMV_results_list_svd_filter.png
    

  Explore spatiotemporal dynamics of beamformer outputs in the right condition
  
  DatabaseExpoloreRightAvgEig.png
  

  • 17.5 ms (left): S1 left 37.5 ms (middle): S1 left

    75 ms (right): bilateral S2
    
    RightEig17p5.png RightEig37p5.png RightEig75.png
    
    

  Explore spatiotemporal dynamics of beamformer outputs in the left condition
  
  DatabaseExplorerLeftAvgEig.png
  

  • 18.3 ms (top-left): S1 right 34.1 ms (top-right): S1 right 60 ms (bottom-left): S2 right

    75.8 ms (bottom-right): bilateral S2
    
    LeftEig18p3.png LeftEig34p1.png
    
    LeftEig60.png LeftEig75p8.png
    
    

Maximum constrast beamformer Maximum constrast beamformer (MCB) is scalar-type beamformer [Chen et al., 2006]. For each position

latex error! exitcode was 2 (signal 0), transscript follows:

, it provides an analytical solution of dipole orientation

latex error! exitcode was 2 (signal 0), transscript follows:

, which maximizes the constrast of beamformer outputs between active state and control state as bellow:

latex error! exitcode was 2 (signal 0), transscript follows:

where

latex error! exitcode was 2 (signal 0), transscript follows:

is the covariance matrix computed from the measurements during active state and

latex error! exitcode was 2 (signal 0), transscript follows:

is the covariance matrix computed from the measurements during control state. Solution of dipole orientation The solution of spatial filter can be rewritten as

latex error! exitcode was 2 (signal 0), transscript follows:

where both

latex error! exitcode was 2 (signal 0), transscript follows:

and

latex error! exitcode was 2 (signal 0), transscript follows:

depend only on the dipole location

latex error! exitcode was 2 (signal 0), transscript follows:

. Then the objective function for obtaining the dipole orientation

latex error! exitcode was 2 (signal 0), transscript follows:

can be rewritten as

latex error! exitcode was 2 (signal 0), transscript follows:

where both of the 3-by-3 matrices

latex error! exitcode was 2 (signal 0), transscript follows:

and

latex error! exitcode was 2 (signal 0), transscript follows:

does not depend on the dipole orientation

latex error! exitcode was 2 (signal 0), transscript follows:

. The solution of dipole orientation

latex error! exitcode was 2 (signal 0), transscript follows:

is the eigenvector corresponding to the maximum eigenvalue of the matrix

latex error! exitcode was 2 (signal 0), transscript follows:

. Statistical mapping After obtaining the dipole orientation

latex error! exitcode was 2 (signal 0), transscript follows:

and the spatial filter

latex error! exitcode was 2 (signal 0), transscript follows:

for dipole position

latex error! exitcode was 2 (signal 0), transscript follows:

, the F-statistic value at time

latex error! exitcode was 2 (signal 0), transscript follows:

can be obtained using the following formula:

latex error! exitcode was 2 (signal 0), transscript follows:

where

latex error! exitcode was 2 (signal 0), transscript follows:

is the covariance matrix computed from the measurements during window

latex error! exitcode was 2 (signal 0), transscript follows:

,

latex error! exitcode was 2 (signal 0), transscript follows:

is the size of

latex error! exitcode was 2 (signal 0), transscript follows:

, and the range of

latex error! exitcode was 2 (signal 0), transscript follows:

is

latex error! exitcode was 2 (signal 0), transscript follows:

.

latex error! exitcode was 2 (signal 0), transscript follows:

is a segment of active state

latex error! exitcode was 2 (signal 0), transscript follows:

. The meaning of F-statistic is the same as the normalized variance

latex error! exitcode was 2 (signal 0), transscript follows:

used in LCMV beamforming method. If the number of

latex error! exitcode was 2 (signal 0), transscript follows:

is small, the peak of temporal dynamics of F-statistic value may shift. In this case, it is more appropriate to replace the covariance matrix

latex error! exitcode was 2 (signal 0), transscript follows:

with

latex error! exitcode was 2 (signal 0), transscript follows:

or

latex error! exitcode was 2 (signal 0), transscript follows:

where

latex error! exitcode was 2 (signal 0), transscript follows:

represents the window of baseline and its size is

latex error! exitcode was 2 (signal 0), transscript follows:

. When

latex error! exitcode was 2 (signal 0), transscript follows:

is used, the F-statistic value represents the estimate of source power to noise variance. MCB process

• Drag and drop all of the epochs in Subject01 / right / right | timeoffset (98 files) into Process1.

  Click on [RUN] and select the process "Sources > Maximum Contrast Beamformer".
  
  MCBPipelineEditorFmap2.png With this window you can select the dipole orientation you want to use, the time window of active state, the window size you are going to use for the computation of F-statistics, the regularization parameter, and the sensors you are going to use for this estimation. You can edit the following options:

  • Comment: This field contains what is going to be displayed in the database explorer.
  • Dipole orientation: Please select Unconstrained, which calculates the dipole orientation using the maximum contrast criterion. The method Cortially Constrained uses the orientation of cortical surface as dipole orientation. Note that the method Cortially Constrained is only available when the Head Model provides the orientation for each grid location. The unconstrained method may give more smooth results than the cortically constrained method.

    Time windows: Three parameters have to be set: Time range of active window, F statistic window size, and Temporal resolution. The dipole orientation and spatial filter is estimated using the variance of source activity during active state

    latex error! exitcode was 2 (signal 0), transscript follows:
    
    

    . F-statistic window size defines the length of the window for the computation of F-statistics, that is,

    latex error! exitcode was 2 (signal 0), transscript follows:
    
    

    . The time window for computing F-statistics is going to be slided in Time range of active state with the interval set to be Temporal resolution. If there are more than one F-statistic windows, the time label of the F statistics will be set as the middle time of each window, that is,

    latex error! exitcode was 2 (signal 0), transscript follows:
    
    

    in the following figure. And you will obtain the spatiotemporal F-statistics bewteen

    latex error! exitcode was 2 (signal 0), transscript follows:
    
    

    and

    latex error! exitcode was 2 (signal 0), transscript follows:
    
    

    .
    
    MCBSlidingWindow.png Regularization parameter: This value will be multiplied with the maximum eigenvalue of the covariance matrix computed from the recordings during Time range of interest. The larger value of regularization parameter may give the more smooth results. Sensors type: Modalities that are used for the reconstruction. Here we only have one type of MEG sensors, so nothing to change. Method for F statistic computation: Please select Variance (use user define baseline) / Variance to compute F-statistic values. The other two methods are Power / Variance and Variance (use user defined baseline) / Variance. When the methiod Variance (use user defined baseline) / Variance is selected, the field of Baseline time window for F-ststistic calculation must be given. In this example, we set the baseline to be -104.2 to -10 ms. Click on [Run].
  Two new files are available in the database explorer.

  • right:MCB: SpatilFilter(Unconstr)|0.0_295.8ms: This file saved the spatial filter for each grid point. The spatial filters are normalized using the standard deviation of beamformer outpouts during control state.
    
    MCBSF.png

  • right:MCB: F-statistics var(b)/var(Unconstr)|0.0_295.8ms: This file saved the F-statistic value for each grid point during 0-295.8 ms.
    
    MCBDatabaseExplorerFmaps.png
    
    Double click on this file. Set the time to be anytime between 0 and 295.8 ms. Then click Surface tab on the right panel of database explorer. Set Data options > Amplitude to be 95%.
    
    MCBFmaps3Methods.png
    
    From left to right, the figures display the F statistics computed using Power / Variance, Variance / Variance, and Variance (use user defined baseline) / Variance. The baseline used in this example was set to be -104.2 ~ -10 ms.

  Process Subject01 / left / left | timeoffset (101 files) using the same procedure.
  
  MCBFmapsleft.png
  
  From left to right, the figures display the F statistics computed using Power / Variance, Variance / Variance, and Variance (use user defined baseline) / Variance. The baseline used in this example was set to be -104.2 ~ -10 ms.
  
  Process Subject01 / right / right | timeoffset (98 files) using the following parameter values to obtain the spatiotemporal dynamics of F-statistics.
  
  MCBPipelineEditorFDynamics2.png
  
  Two new files are available in Database explorer:

  • right:MCB: SpatilFilter(Unconstr)|0.0_295.8ms: This file saves the spatial filter for each grid points. The filename shows the time range used to compute spatial filters, that is, 0.0 to 295.8 ms in this case. right:MCB: F-statistics var(b)/var(Unconstr)|10.0_290.0ms(ws:20.0ms,tr:10.0ms,bs:-104.2_-10.0ms): This file saved the spatiotemporal dynamics of F-statistics between 10.0 and 290.0 ms. The abbreviations for F-statistics window size,temporal resolution, and baseline is ws, tr, and bs. Click on this file. Choose Surface tab and set Data options > Amplitude to be 50%. Right click on the pop-out window. Select Snapshot > Time contact sheet: Figure.
    
    MCBSnapshotSettting.png
    
    In this window, set Start time (ms) to be 10, End time (ms) to be 290, and Number of snapshots to be 29. Click on Ok.You will obtain the following figures.
    
    MCBFdynamicsRight.png
    
    These figures display the F-statistics maps between 10 ms and 290 ms with 10-ms time interval. The baseline used in this example is set to be -104.2 ~ -10 ms.

  Process "Subject01 / left / left | timeoffset (101 files)" with the same parameter values used in the previous step to obtain the spatiotemporal dynamics of F-statistics.
  
  MCBFdynamicsLeft.png

Beamformer-based correlation/coherence imaging Dynamic imaging of coherent sources (DICS) [Under construction] Spatiotemporal imaging of linearly-related source components (SILSC) [Under construction] References

• Liu TC and Van Veen BD, (1992)
  Multiple window based minimum variance spectrum estimation for multidimensional random fields
  IEEE Transactions on Signal Processing, 40(3): 578-589 Van Veen BD, Van Drongelen W, Yuchtman M, Suzuki A (1997)
  Localization of brain electrical activity via linearly constrained minimum variance spatial filtering
  IEEE Transactions on Biomedical Engineering 44(9): 867-880. Gross J, Kujala J, Hamalainen M, Timmermann L, Schnitzler A, Salmelin R (2001)
  Dynamic imaging of coherent sources: Studying neural interactions in the human brain
  Proc Natl Acad Sci U S A 98(2): 694-699. Robinson SE and Vrba J (1999)
  Functional neuroimaging by synthetic aperture magnetometry
  Recent Advances in Biomagnetism. Tohoku University Press 1999: 302-305. Chen YS, Cheng CY, Hsieh JC, Chen LF (2006)
  Maximum contrast beamformer for electromagnetic mapping of brain activity
  IEEE Transactions on Biomedical Engineering 53(9): 1765-1774.

Feedback

---