Hi, Ellen-
For diffusion-based matrices (*.grid files, from 3dTrackID), each matrix element should be >0 if a tract was found, and only 0 if a tract was not found. It would make sense that those could then be treated as NA in R.
For FMRI-based matrices (*.netcc files, from 3dNetCorr), it is *possible* that a matrix element could be exactly zero from correlation (though, veeery small but nonzero is more likely for cases of being highly uncorrelated). So I think you would basically always want all FMRI elements, even the zeros.
--pt