I know that there is ~10% segmentation error rate in the Microns/Neuroglancer Layer 2/3 volume. Is there a place or centralized way to report segmentation errors that will help improve the fidelity of the dataset over time?
As an example, I was looking at the synapse table (pni_synapses_v185) and sorting for cleft_vx to find the largest cleft size. When I revealed the synapse in Neuroglancer, it does indeed seem to be a large cleft volume, however, the synapse ID in neuroglancer for this synapse did not match the synapse ID in the synapse table. In other words, the synapse table seemed to be correct for everything except the synapse ID in this particular instance.
I followed the rabbit trail to see where the erroneous ID was located in the synapse table, and it led to another synapse, again which had a different synapse ID when viewed in Neuroglancer. Finally, I looked up the third erroneous synapse ID and such a synapse ID did not exist in the synapse table.
I put a Jupyter notebook file here on GitHub showing this:
With a 10% error rate, segmentation errors will come up somewhat often, so I just expect to come across errors now and again. Is there a process for updating the v185 dataset and any process for contributing to improving the dataset? Thanks,
I just now realize one potential reason for this mismatch; in the pni_synapses_v185 table; there is an ID column, which I had assumed was the synapse ID, but perhaps this is not the case. Maybe the ID column is simply an index number and not the synapse ID. That might explain why the ID in the table does not match with the synapse ID in neuroglancer.
So, I guess I need to update this thread to have two questions:
- Clarification on where to find the synapse ID in the pni_synapses_v185 table; and
- The original question: is there a means to submit segmentation errors to improve the fidelity of the v185 dataset over time?
You’re correct about the reason for the mismatch. The ID column in that table represents a different ID from the neuroglancer ID. We’ve generated a version of that file with the neuroglancer ID in that column instead and should replace it on the Binder website soon.
Thanks, I will keep an eye out for the updated table! Do you know if it will that appear on the Microns Explorer page too (that’s where the current synapse table resides)?
Any thoughts on processes for improving the segmentation fidelity over time? Is there a ‘curator’ or centralized way to update this volume?
Hi Shawn, i just wanted to let you know that i updated the synapse table that is linked to on microns explorer to a new version that has the IDs that match. I also added the nucleus id to the soma valence table so you can lookup the nucleus of any the seg_ids that have nuclei.
In terms of improvements we do have a way of making improvements in the segmentation, if you’d like to do that work, please email me at forrestc at alleninstitute dot org and we’ll arrange a discussion/tutorial about how to get access and do that work.
Thank you, I very much appreciate your time and help! I’m very interested in being part of the effort. I will follow up separately by email. Best regards,