CCFv3 annotation: wrong IDs

Dear forum users,

I heard an image with the ID of annotation as intensity, so I downloaded it at the following URL.

http://download.alleninstitute.org/informatics-archive/current-release/mouse_ccf/annotation/ccf_2017/

The other nrrd files did not seem to have the ID intensity, but I thought annotation_10.nrrd matched the ID intensity at first glance. However, for example, there is a region with an intensity of 606826624, and the supplementary information table in the Cell paper (Wang et.al Cell 2020) shows that there are regions with similar IDs, but none of them match exactly. It seems that there are a number of regions with intensities that do not match the atlas ID. Where can I download the nrrd file of the annotation with the correct ID?

Thanks in advance for your answer.

Best regards,
Tomo

@tomo

First let’s confirm that you are referring to TableS2 of the paper (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8152789/)

The values you see in the table was generated directly from annotation_10.nrrd. There shouldn’t be any intensity in that file that is not in that supplementary table.

In each row of the table, the total_voxel_counts is obtain by summing not just the label of the structure of interest but also all its children.

How are you opening the file? Is it keeping it as uint32 format?
If you are using python pynrrd is known to work

If you continue to have difficulty with the uint32 file, the other option is to access the structure through AllenSDK Reference Space — Allen SDK dev documentation

Could you give these are try?

@lydian
Thank you very much for your reply.
I used uint32 format and pynrrd to manage the annotation images, but it seems to be a problem with ImageJ. I made a tiff file of the annotation image, then read it and looked at the numpy array data, and the wrong ID was not caught.
After converting the large ID to a normal value that I decided on, I was able to display the image correctly on ImageJ.

I think I have solved the problem for now.
Thank you very much for your help.

There have been issues with large ID and ImageJ. The nrrd image is converted to float 32 bits in ImageJ, which leads to a precision loss for large IDs. For instance 606826624 and 606826625 will be both converted to the same 6.068226e8 value, resulting in indistinguishable labels.

I am curious though: why were such big label numbers chosen, given that there are way less than 2e16 labels in the atlas ? Is there a particular meaning to these big numbers ?