[ENH]: Add non zero mean as an aggregation function for structural images

[antogeo]

Are you requiring a new dataset or marker?

• I understand this is not a marker or dataset request

Which feature do you want to include?

How do you imagine this integrated in junifer?

additional parameter

Do you have a sample code that implements this outside of junifer?

numpy.mean(numpy.nonzero(a))

Anything else to say?

no

[fraimondo]

Can you provide more information as requested (e.g. by properly filling the issue template)?

np.nonzero gives the indexes of the non-zero elements.

Also, doing a non-zero mean of floating point values will only work if values are exactly 0.

[antogeo]

0 values indicate those voxels that despite belonging to a parcel they are certainly not part of tissue of interest. Therefore, should not be involved in the mean process. A cut-off value could also be a more general solution (eg a threshold applied in the tissue of interest).

[LeSasse]

0 values indicate those voxels that despite belonging to a parcel they are certainly not part of tissue of interest. Therefore, should not be involved in the mean process. A cut-off value could also be a more general solution (eg a threshold applied in the tissue of interest).

By 0 values, do you mean voxels where the actual data is 0 (or below some threshold) or do you mean voxels where they are 0 (or below some threshold) in an additional probabilistic mask?

[fraimondo]

I still do not get how this would be preferred instead of using a GM mask or something like that.

It's just basically thresholding whatever values you will aggregate.

How will this work with timeseries?