I am working with a brain tumor dataset - because BOLD signal from lesioned voxels can be unreliable (e.g., due to vascular uncoupling), my analysis is restricted to the contralesional hemisphere. However, the voxelwise analysis used to calculate DOF and QC measures is from the whole brain, which could include lesioned voxels. Is it possible this is contributing to my DQ score of ~80% (my DV and DS scores are both > 95%)? If I were to mask out the lesion files from the functional dataset, would the QC analyses still be built off 1000 nonzero voxels, and if so, do you recommend this approach?

Also, some of the patients with more severe pathology have a low proportion of valid scans (~38-50%). I multiplied the number of valid scans by the TR /60 to get the total minutes of valid scanning time, and a few patients have < 5 minutes. As expected, these patients often also have low DOF (<20). Do you have guidance for participant exclusion? I am torn as sysematically excluding the sickest subjects introduces bias but I also want to make sure the FC data are reliable! If it matters, I preprocessed the fmri data in fmriprep so the scrubbing/realignment values are calculated using fmriprep's defaults. 

Many thanks!