<div dir="ltr"><div><div>  Dear community,<br><br>  My name is Jan I am based at Nicolaus Copernicus University in Toruń<br>  (Poland), where I study CogSci, I am also member of small research<br>  group focused on on neuronal activity indices and reconstruction<br>  methods.  FieldTrip has turned out to be the most convenient and<br>  capable toolbox for the work we have done so far :)<br><br>  However, we have a question in respect of forward modelling.  We had<br>  a close look into some properties of leadfields and we are anxious<br>  if the features revealed are unusual, alarming or just standard and<br>  irrelevant?<br><br>  We have found that some leadfield nodes have norm (calculations<br>  described below) tremendously higher than the neighbouring nodes.<br>  Also some nodes are column-wise highly correlated.<br><br>  We obtained very similar results for both =icosahedron642= and<br>  random =sphere-like= triangulations.<br><br>  We have produced 2 volume conduction models and 2 corresponding<br>  leadfields that were based on =icosahedron642= or random<br>  =sphere-like meshes= geometry.  Radia for { 'brain' 'skull' 'scalp'<br>  } were [ 88 92 100 ].<br><br>  Electrodes (162, =icosahedron162=) were placed uniformly around the<br>  scalp All geometrical units were in mm.  Conductivity was expressed<br>  in in S/mm.<br><br>  We used =dipoli= and =openmeeg= methods with ft_prepare_headmodel<br><b><span style="font-family:monospace,monospace"><br>  #+BEGIN_SRC matlab :eval no :exports code<br>    cfg                 = [];<br>    cfg.method          = 'openmeeg';<br>    cfg.conductivity    = sel_msh02.cond<br>    sel_vol02_openmeeg  = ft_prepare_headmodel( cfg, sel_msh02.bnd );<br>  #+END_SRC</span></b><br><br>  and<br><br><b><span style="font-family:monospace,monospace">  #+BEGIN_SRC matlab :eval no :exports code<br>    cfg                 = [];<br>    cfg.method          = 'dipoli';<br>    cfg.conductivity    = sel_msh02.cond<br>    sel_vol02_dipoli    = ft_prepare_headmodel( cfg, sel_msh02.bnd );<br>  #+END_SRC</span></b><br><br>  The laedfield was created using the following settings<br><br><b><span style="font-family:monospace,monospace">  #+BEGIN_SRC matlab :eval no :exports code<br>    cfg                   = [];<br>    cfg.elec              = sel_elec00;<br>    cfg.grid.unit         = 'mm';<br>    cfg.grid.xgrid        = -110:5:110;       % Specify in mm!<br>    cfg.grid.ygrid        = -110:5:110;       % Specify in mm!<br>    cfg.grid.zgrid        = -110:5:110;       % Specify in mm!<br>    cfg.reducerank        = 'no';<br><br>    cfg.vol               = sel_vol02_dipoli;<br>    sel_lfg02_dipoli      = ft_prepare_leadfield(cfg);<br><br>    cfg.vol               = sel_vol02_openmeeg;<br>    sel_lfg02_openmeeg    = ft_prepare_leadfield(cfg);<br>  #+END_SRC</span></b><br><br>** Norm<br><br>   Next, we plotted norm of the leadfields expressed as<br><br>   $norm(H) = sum(sum(abs(H)))$<br><br>   [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_712_leadfield-norms_openmeeg.png" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_712_leadfield-norms_openmeeg.png</a>]]<br><br>   On the figure above norm for OPENMEEG created leadfield is<br>   expressed by colour and marker size.  It is quite evident that<br>   leadfields with very high norm are distributed on the externals of<br>   the grid.<br><br>   [[file:img/fig_711_leadfield-norms_dipoli.png]]<br><br>   Same problem arises with DIPOLI leadfield, but to much smaller<br>   extent.  My it pose a serious threat to both spatial filters and<br>   some activity indices (such as power of LCMV filter).<br><br>   The following histograms also show the same feature of the<br>   leadfields.  Please note that the horizontal axis contains<br>   log(norm(H)) so the values spread is more dramatic than it appears<br>   at the first glace.<br><br>   [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_722_leadfield-normHist_openmeeg.png" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_722_leadfield-normHist_openmeeg.png</a>]]<br><br>   [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_721_leadfield-normHist_dipoli.png" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_721_leadfield-normHist_dipoli.png</a>]]<br><br>** Correlation<br><br>   We also investigated correlation between columns of each leadfield<br>   node.  We suspect that this feature can also have deleterious efect<br>   on performance of some spatial filters and neuronal activity<br>   indices.<br><br>   Following scatterplots show this using color and marker size which<br>   reflect the absolute value of correlation coefficient between<br>   second and third column of each leadfield node (change of columns<br>   in consideration does not alleviate the problem.<br><br>   OPENMEEG<br>   [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_732_leadfield-corrCoefs_openmeeg.png" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_732_leadfield-corrCoefs_openmeeg.png</a>]]<br><br>   DIPOLI<br>   [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_731_leadfield-corrCoefs_dipoli.png" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_731_leadfield-corrCoefs_dipoli.png</a>]]<br><br>** Differences between leadfields<br><br>   Futhermore we checked correlation between solutions provided by<br>   DIPOLI and OPENMEEG.  Here we calculated correlation coefficients<br>   between the two leadfield grids with respect to nodes of the same<br>   position inside the ``brain''.<br><br>   Following plot shows only leadfield nodes for which the absolute<br>   value of correlation coefficient was lower than 0.1.<br><br>   This shows that solutions for the nodes that are located deep<br>   inside brain are highly correlated.  The main differences between<br>   the leadfields are distributed at the exterior of the grid.<br><br>   [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_800_leadfield-corrCoefs-DIPOLI-vs-OPENMEEG.png" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_800_leadfield-corrCoefs-DIPOLI-vs-OPENMEEG.png</a>]]<br><br>** MATLAB figures<br><br>   * [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_666_triangulation-meshes.fig" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_666_triangulation-meshes.fig</a>]]<br>   * [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_700_electrode-positions-and-labels.fig" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_700_electrode-positions-and-labels.fig</a>]]<br>   * [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_701_leadfield-geometry_dipoli.fig" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_701_leadfield-geometry_dipoli.fig</a>]]<br>   * [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_702_leadfield-geometry_openmeeg.fig" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_702_leadfield-geometry_openmeeg.fig</a>]]<br>   * [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_711_leadfield-norms_dipoli.fig" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_711_leadfield-norms_dipoli.fig</a>]]<br>   * [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_712_leadfield-norms_openmeeg.fig" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_712_leadfield-norms_openmeeg.fig</a>]]<br>   * [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_721_leadfield-normHist_dipoli.fig" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_721_leadfield-normHist_dipoli.fig</a>]]<br>   * [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_722_leadfield-normHist_openmeeg.fig" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_722_leadfield-normHist_openmeeg.fig</a>]]<br>   * [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_731_leadfield-corrCoefs_dipoli.fig" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_731_leadfield-corrCoefs_dipoli.fig</a>]]<br>   * [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_732_leadfield-corrCoefs_openmeeg.fig" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_732_leadfield-corrCoefs_openmeeg.fig</a>]]<br>   * [[<a href="https://github.com/nikadon/lfgTesting001/blob/master/img/fig_800_leadfield-corrCoefs-DIPOLI-vs-OPENMEEG.fig" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/img/fig_800_leadfield-corrCoefs-DIPOLI-vs-OPENMEEG.fig</a>]]<br><br>   Details of the MATLAB implementation for the above can be found on<br>   <a href="https://github.com/nikadon/lfgTesting001/blob/master/lfgTesting001.org" target="_blank">https://github.com/nikadon/lfgTesting001/blob/master/lfgTesting001.org</a><br><br><br>   Thank you in advance for any help or comments...<br><br>   Best,<br>   Jan<br></div></div><br></div>