Julich Brain 3.0 and 3.1 (prerelease) have some PMAPs that are not masked with the templates

[xgui3783]

as of 0.4a63

sample code:

import siibra

pt_spec = siibra.Point.parse("18.505544mm,-1.754904mm,33.293708mm")
pt = siibra.Point(pt_spec, space=siibra.spaces['mni152'])
statmap = siibra.get_map("3.0.3", "mni152", "statistical")
df = statmap.assign(pt)
print(len(df)) # expects 0, got 1
print(df['region']) # expects None, got `Ventral Dentate Nucleus (Cerebellum) right`

atlas location clearly does not point to any region

related https://github.com/FZJ-INM1-BDA/siibra-explorer/issues/1401

[AhmetNSimsek]

The labelled map indeed returns empty. With statistical map, the map value is 0.0. The assignment should not have passed as the threshold is 0.0. I will investigate.

[AhmetNSimsek]

The assignmnet values is 2.6801472152726546e-09 but this shows up as zero (I think due to relative tolerance). We can remedy this by changing the threshold from 0.0 to sth more reasonable as the default value. What do you think @xgui3783 and @dickscheid ?

[AhmetNSimsek]

Any non-zero assignment is an assignment and user can utilize a different threshold if they wish to. Perhaps siibra-explorer can call this with a different threshold, @xgui3783 ?

[xgui3783]

We can certainly use a preset threshold.

But what should we use? 0.05? I heard from statistian friends that that's usually significant enough (/joke)

[AhmetNSimsek]

It seems only 4 digits are visible after the comma. I'd say then at least 0.0001. But I think Timo should have the final say. (There is a point in the agenda already, we can clarify this week.)

[xgui3783]

As discussed on developmeent meeting on 2024-01-17, this issue should be addressed the following way:

• [x] the code/original probabilistic map should be double checked that at specific MNI coordinate, the assignment is faithful to the original source data
• [ ] if so, the source data should be corrected

[AhmetNSimsek]

It is indeed correct. I passed the info forward to Julich Brain team. As the code is functioning correctly and it is in fact a data issue (if it is an issue, they will tell me as well), I am transferring this to siibra-configurations.

[dickscheid]

I agree it should be solved on the data side. Using a default threshold is difficult, since the interpretation of values is different for different maps (e.g. difumo vs. julich-brain). Of course it should be possible to use a threshold, but it needs to be explicit, not implicit then. If no threshold is given, which should be the default in my eyes, the dataset (parcellation map) defines which regions are assigned with its nonzero values. Of course, the given example is anything else than intuitive and should be clarified.

[AhmetNSimsek]

I've just talked with Hartmut who confirmed it is an error. He will check his original data and contact us.

[AhmetNSimsek]

I've just talked with Hartmut who confirmed it is an error. He will check his original data and contact us.

Hartmut said it is indeed a data issue at the source. This will be remedied for Julich Brain 3.1.

[AhmetNSimsek]

It is remedied with full release of 3.1. 3.0 cannot be fixed since it is released already and there is a new minor version (3.1)