I've run LPA supplying both a T1-w and FLAIR image for WMH segmentation on an AD dataset. I see NaNs appearing in several voxels after inspecting outputs and am confused by what they mean. From the code, it looks like NaNs aren't necessarily "0" voxels, but they also aren't necessarily WMHs either (especially since some of the NaNs I inspect in FSLEyes capture some large WMHs). I am wondering if there is an approach that might appropriately deal with NaNs - set them to 0 in cases where it's unlikely to be a WMH, but give them a non-zero value in cases where it's more likely to be a WMH?

Any assistance would be greatly appreciated!