For most studies it is necessary to reduce the effect of non-task-related variance in the data in order to increase the sensitivity of the statistical analysis and to detect effects of interest. Such sources of noise variance include:

• system noise, as for example signal drifts
• physiological noise, including head and body motion as well as respiration and cardiac effects
• and noise due to stimulation of no interest, e.g. auditory stimulation because of scanner noise

Therefore, often numerous confounds are included in design matrices, aiming to improve the fit of the model to the data by accounting for noise variance. As the number of included predictors in the design matrix increases the number of degrees of freedom in the analysis decreases. Hence, it is worthwhile to investigate whether there is an added advantage of including more confounds in a model. The effectiveness in explaining additional noise variance in the data by including more confound predictors can be investigated with the plugin at hand. Two different models (SDMs) are compared and it is assumed that SDM1 is nested within SDM2, i.e. SDM2 contains the same task and confound predictors as SDM1 as well as additional confounds that are not included in SDM1.

• a functional data set has to be open in BrainVoyager (e.g. test.fmr)
• two models defined for the active FMR have to be loaded in the plugin with SDM1 being nested within SDM2 (e.g. SDM1: task.sdm and SDM2: task_3DMC.sdm)

1. Computation of the temporal signal-to-noise ratio (tSNR) of the data after cleaning (removing the confound variance in the data)
   Option: '(1) tSNR residuals of SDM1 confounds', '(2) tSNR residuals of SDM2 confounds'
   The two models defined in the fields 'SDM1' and 'SDM2' will be fitted to the data of the active FMR project. The residual time courses, resulting from subtracting the confound predictor time courses x their betas from the data time course, will be used to calculate the tSNR of the data after cleaning. The tSNR is defined as the average signal intensity of the time course divided by its standard deviation.
• example output:
• task_tSNR_residuals.map. This is the tSNR of the residual FMR after removing confound variance from SDM1: task.sdm. Since there were no confounds included in SDM1 for this example, the tSNR of the residual FMR is the same as the tSNR of the input FMR.
• task_3DMC_tSNR_residuals.map. This is the tSNR of the residual FMR after removing confound variance from SDM2: task_3DMC.sdm.


2. Comparison of the tSNR of the data using the confounds in SDM2 versus SDM1 for cleaning
   Option: '(3) tSNR difference: (2) - (1)'
• example output:
• task_3DMC_tSNR_diff_task.map.


3. Computation of the explained variance in the data (adjusted R-squared)
   Option: '(4) R2 adjusted SDM1', '(5) R2 adjusted SDM2'
   SDM1 and SDM2 will be used to calculate a GLM on the data of the active FMR project. The explained variance in the data, adjusted for the number of included model predictors (adjusted R-squared), will be computed for SDM1 and SDM2 and saved in two separate maps.
• The adjusted R-squared is defined as follows:
• R-squared = 1 - ( SSres / SStot )
• Adjusted R-squared = 1 - ( 1 - R-squared ) * ( ( N - 1 ) / ( N - P - 1 ) )
  where SSres is the residual sum of squares, SStot is the total sum of squares, N is the sample size and P is the number of predictors in the SDM
• example output:
• task_R2adj.map. This is the adjusted R-squared map for the GLM computation with SDM1: task.sdm.
• task_3DMC_R2adj.map. This is the adjusted R-squared map for the GLM computation with SDM2: task_3DMC.sdm.


4. Comparison of the explained variance in the data using SDM2 versus SDM1 for the GLM computation
   Option: '(6) R2 adjusted difference (5) - (4)'
• example output:
• task_3DMC_R2adj_diff_task.map.


5. The improvement in the resulting GLM using SDM2 versus SDM1 assessed by an F-map
   Option: '(7) F map SDM2 vs SDM1'
   The improvement in explaining the data by using the more complex model (SDM2) is assessed via an F-map.
   • The F-map is computed as follows:
   • F = ( ( SSres_SDM1 - SSres_SDM2 ) / ( P_SDM2 - P_SDM1 ) ) / ( SSres_SDM2 / ( N - P_SDM2 ) )
   • example output:
   • task_3DMC_Fval_diff_task.map.
   
   
6. Saving the cleaned data to disk (saving the residual FMR after confound cleaning)
   Checkmarks: 'FMR residuals after confound cleaning SDM1', 'FMR residuals after confound cleaning SDM2'
   The residual time courses of the data after removing the confound variance from SDM1/SDM2 will be saved to disk.
   Please note that you might remove variance of interest, if your confound predictors are highly correlated with your task predictors!
• example output:
• test_min_confounds_of_task.fmr. This is the residual FMR after removing confound variance from SDM1: task.sdm. Since there were no confounds included in SDM1 for this example, the residual FMR is the same as the input FMR.
• test_min_confounds_of_task_3DMC.fmr. This is the residual FMR after removing confound variance from SDM2: task_3DMC.sdm.