Request: continue producing summary statistics and associated measures from unweighted values

[psadil]

What would you like to see added in this software?

Possibly related to #1383

We recently upgraded MRIQC from 0.15.1 -> 24.0.2, and the distribution of many values derived from summary statistics changed substantially. We noticed this because we closely track the CNR of anatomicals.

Not sure, but this appears to be the relevant commit. From what I'm seeing, the commit message ("enh: deep revision of the summary statistics calculation") did not make it into any release notes (based on dates, I think it would have been between release 23.0.1 and 23.1.0).

Here's how I understand the difference

• before: the summary stats (e.g., mean WM intensity, std of WM) were calculated with the raw signal values of all voxels in thresholded versions of the probabalistic mask (all voxels where prob > 0.85)
• now: the summary stats are calculated with all voxels, weighting by the probability mask.

I assume that the older method (based on thresholded segmentations) is the more common approach in other pipelines (and that older method seems better matched to the documentation, e.g., for CNR). To facilitate interpretation and comparisons with distributions from other pipelines, it would be nice to continue generating summary stats with the older method (and derived measures like CNR).

The new, weighted approach seems like it could also produce an interesting set of metrics. If they're to be produced, it would be nice if they were an addition rather than replacement (e.g., with the suffix _weighted)

Do you have any interest in helping implement the feature?

Yes

Additional information / screenshots

No response

[psadil]

Just wanted to follow-up on this to show how different the values are between releases