I updated the StitchVolumes module to respect soft-applied transforms in this PR: update to respect soft-applied transforms by mikebind · Pull Request #15 · PerkLab/SlicerSandbox · GitHub.
Blending multiple overlapping image volumes where the overlaps could be in arbitrary directions is much more complicated, and I don’t have time to implement it now, but I will make a note of it and perhaps be able to implement it in the future. I realize that doesn’t solve your current problem, @Young_Wang, but perhaps you can take a look at the existing code for the module for inspiration about how to best approach what you would like to accomplish. Ideas which could make sense to me would be resampling to a common voxel grid, and then perhaps doing some type of thresholded image math to assemble the final image volume.
I don’t clearly understand where your data comes from or what it looks like (particularly in 3D), but if you want to start in this direction and can share some more about your problem I’d be happy to try to offer more specific suggestions.