small updates

exercises_code/ni2_inverse5_scanning.m

@@ -97,6 +97,7 @@
 cfg.headmodel  = headmodel;
 % cfg.normalize = 'yes';
 sourcemodel = ft_prepare_leadfield(cfg);
+
 L = cat(2,sourcemodel.leadfield{sourcemodel.inside});
 
 % compute the covariance
@@ -236,41 +237,45 @@
 cfg.funparameter = 'avg.pow';
 cfg.method = 'slice';
 cfg.nslices = 10;
-cfg.funcolorlim=[0 0.2];
+cfg.funcolorlim = [0 0.2];
 ft_sourceplot(cfg,source);
 
 %% contrast between 2 conditions
 
 % create the sensor data for the second condition
 sensordata2 = 1.25.*leadfield1*s1 + 0.8.*leadfield2*s2 + 0.8.*leadfield3*s3 + 1.25.*leadfield4*s4 + randn(301,1000)*0.04e-8;
 
-% compute the covariance
-C2  = cov([sensordata sensordata2]');
-iC2  = pinv(C2); % so that we compute it only once
+% compute the covariance and its inverse for both conditions combined
+C2   = cov([sensordata sensordata2]');
+iC2  = inv(C2);
 iCr2 = inv(C2+eye(301)*1e-19);
 
-% compute the beamformer spatial filter
+% compute the beamformer spatial filter for condition 2
 for ii = 1:size(L,2)/3
   indx=(ii-1)*3+(1:3);
   Lr = L(:,indx);  % Lr is the leadfield for source r
   wbfr2(indx,:)=pinv(Lr'*iCr2*Lr)*Lr'*iCr2;
 end
-sbfr2 = wbfr2*sensordata2;
-pbfr2 = var(sbfr2,[],2);
-pbfr2 = sum(reshape(pbfr2,3,[]));
+
 sbfr1 = wbfr2*sensordata;
 pbfr1 = var(sbfr1,[],2);
 pbfr1 = sum(reshape(pbfr1,3,[]));
+
+sbfr2 = wbfr2*sensordata2;
+pbfr2 = var(sbfr2,[],2);
+pbfr2 = sum(reshape(pbfr2,3,[]));
+
 source.avg.pow(source.inside)=(pbfr1-pbfr2)./(pbfr1+pbfr2);
 
 cfg = [];
 cfg.funparameter='avg.pow';
-cfg.method='slice';
+cfg.method='ortho';
 cfg.nslices = 10;
 cfg.funcolorlim=[-.2 .2];
 ft_sourceplot(cfg,source);
 
 %% correlated sources
+
 sens = ni2_sensors('type','meg');
 headmodel = ni2_headmodel('type','spherical','nshell',1);
 
@@ -295,6 +300,7 @@
 cfg.vol  = headmodel;
 % cfg.normalize = 'yes';
 sourcemodel = ft_prepare_leadfield(cfg);
+
 L = cat(2,sourcemodel.leadfield{sourcemodel.inside});
 
 % compute the covariance

exercises_text/ni2_inverse5_scanning.md

@@ -304,10 +304,10 @@ We will use the variables `sensordata`, `sourcemodel` and `L` that we also used
 
 Now we will simulate data from a 'second’ condition (compared to the original variable sensordata), where the sources have the exact same locations and time courses, but the amplitude of two sources is decreased, and the amplitude of the other sources is increased, relative to the 'first’ condition.
 
-    sensordata2 = 1.25.*leadfield1_s1 + ...
-                  0.80.*leadfield2_s2 + ...
-                  0.80.*leadfield3_s3 + ...
-                  1.25.*leadfield4_s4 + ...
+    sensordata2 = 1.25 .* leadfield1*s1 + ...
+                  0.80 .* leadfield2*s2 + ...
+                  0.80 .* leadfield3*s3 + ...
+                  1.25 .* leadfield4*s4 + ...
                   randn(301, 1000)*0.04e-8;
 
 We will now compute the spatial filters using the covariance estimated from the data combined across the two conditions. In this way we will end up with a single set of spatial filters, and thus, in comparing across the two conditions, we ensure that the depth bias is the same for condition 1 and 2. The covariance of the data combined across the 2 conditions can be computed in several ways, e.g. by averaging the single condition covariances. Here, we first concatenate the data and subsequently compute the covariance. Now we will use the MATLAB `cov` function, rather than computing the covariance 'by hand’ (i.e. by first computing the mean across time etc.).