SPM Write Residuals

JasminStein · August 24, 2023, 6:58am

Hi everyone,

I want to do some data analysis on the residuals from a model estimation in SPM. However, I have noticed that SPM did not return one residual image per original functional volume but that quite some residuals seem to be missing after model estimation. Does anyone know why this might be the case?

Thanks in advance!
Jasmin

Remi-Gau · August 24, 2023, 9:13am

This seems related to this topic.

JasminStein · August 24, 2023, 9:37am

Thanks, but I know how to make SPM write and keep the residuals; the problem here is that, say my functional run has 600 images, but I am only getting 500 residual images as output from SPM. The missing ones seem pretty randomly distributed amongst the 600 volumes. I wonder why that is?

Remi-Gau · August 24, 2023, 12:40pm

This is strange. Any luck asking on the SPM mailing list?

Guillaume · August 31, 2023, 12:06pm

Hi @JasminStein,

How are you saving the residual images? Are you using the write_residuals option during model estimation as suggested by @Remi-Gau or are you looking at the temporary ResI_????.nii files? With the latter, only a subset of them are saved (64 by default), uniformly sampled over repetitions (time here):

github.com

spm/spm/blob/main/spm_spm.m#L523


      
              VResI(i).fname   = [sprintf('ResI_%04d', i) file_ext];
              VResI(i).descrip = sprintf('spm_spm:ResI (%04d)', i);
          end
          VResI = spm_data_hdr_write(VResI);
          
          
          %==========================================================================
          %-               G E N E R A L   L I N E A R   M O D E L
          %==========================================================================
          
          iRes = round(linspace(1,nScan,nSres))';              % Indices for residual
          
          %-Get explicit mask(s)
          %==========================================================================
          for i = 1:numel(xM.VM)
              if ~(isfield(SPM,'xVol') && isfield(SPM.xVol,'G'))
                  %-Assume it fits entirely in memory
                  C = spm_bsplinc(xM.VM(i), [0 0 0 0 0 0]');
                  v = true(DIM);
                  [x1,x2] = ndgrid(1:DIM(1),1:DIM(2));
                  for x3 = 1:DIM(3)

If you are using the former (or directly calling spm_write_residuals), it would be useful to have more details as this would be a bug.