indexing error, analysis step fails

[viktorpm]

Describe the bug Cellfinder ran with the pre-trained network without any issues. After retraining the network the analysis step failed with the error message: IndexError: index 367 is out of bounds for axis 2 with size 367

The analysis folder with the summary.csv was not created. Classification ran and I was able to visualize the cells.

I'm working with rat data and used the Waxholm Space atlas of the Sprague Dawley rat brain for the registration step. Initially, it was not perfect so I thought that might have been causing the issue but got the same error even with a fairly good registration.

I tried using another trained model (not the best-performing one) but it didn't help. I only have this issue with one animal. It runs on six others with similar parameters without an issue.

Log files cellfinder.out.txt cellfinder.err.txt

[adamltyson]

I think this is a general issue and I suspect the error is something along the lines of:

• The classification model is imperfect and cells are detected outside of the brain OR the registration is imperfect and cells are detected within the brain, but outside of atlas template when registered to the sample
• These "extreme" cells cause the transformation from sample -> atlas space to fail because in some areas of the image there isn't a valid mapping between the two spaces

Two possible solutions:

• Only transform cells that fall within the atlas boundary
• Catch these errors and report the number of "ignored" cells.

[adamltyson]

This is linked to cellfinder-napari PRs https://github.com/brainglobe/cellfinder-napari/pull/117 & https://github.com/brainglobe/cellfinder-napari/pull/153