Distributed module over-merging labels in dense regions

[vbrow29]

I have been using the distributed module to run nuclei segmentation on a large light-sheet microscopy dataset. It mostly works well, but I noticed when quantifying morphological features for each nucleus that I see fewer nuclei centroids in the region between chunks (see below).

I dug a bit and found that the merge-relabeling step tends to over-merge nuclei, especially in dense regions (see before and after relabeling images below):

I've attempted to implement a solution to only merge labels if their centroid-centroid distance is below some threshold and that seems to help, but I'm wondering if there is a better way to go about this. Probably a question for @GFleishman ?

I plan to use this module extensively so I'm open to testing any potential solutions or improvements. Thanks!