Methods for analysis of multisubject studies

Composite Maps

• Methods of calculation
  • Averaged Maps (mean performs better than median)
  • Contrast Maps (use the linear combination of various maps as defined by Contrast
  • General Linear Model Maps (multi-variate approach)
  • Corelation Maps : correlation with additional parameters, age, condition, behavioral score
• assessment of significance
  • no assessment at all - maps used only as a visualization of ROI results
  • assessment based on assuption of character of the noise - t-stat, GLM
  • randomization - computationally intensive and not completely assumption free
    
• Implementation of randomization :
  • We calculate the average map obtained from all the studies with appropriate contrasts.
  • We change sign in half of maps going into composite , this is performed randomly or to exhaust all possible combination.
  • For each voxel the value obtaine in step 1 is compared to randomized (noise) values, p-value is defined as : p < (n+1)/r
    here r is total number of randomizations and n is number of randomizations that exceeded the "real" value.
  The same technique may (and will) be used to test the validity of results obtained using GLM, Correlationm ....