The most recent release for fMRIPrep has the following information in the release log
This release fixes a bug for phase-difference fieldmaps that are not in RAS+ orientation . The bug presented as an error if the orientation was reordered relative to RAS+ (for example, AIL+) and the swapped dimensions were not of the same size. Otherwise, the bug introduced a poor masking of the phase difference map, and could be quite subtle if the original orientation was LAS+.
The wording is confusing to me. It indicates this could be a problem for images if the original orientation was LAS+, but the second sentence indicates it’s a problem if the orientation was reordered relative to RAS+, and LAS+ has the dimensions in the same order.
Is this a problem for any image that is literally not in RAS+ orientation (so the sentence about dimensions reordered is inaccurate) or is it only if dimensions are reodered (and so LAS+ isn’t an issue)?
It is a problem for any phase-difference field map image that is not in RAS+ orientation.
The additional sentences are about how the error would appear. For LAS+ phase-difference field map images, if the images are nearly symmetric left/right, then the error would be very subtle. For an LPI+ image, for instance, we would expect it to be flipped about all axes, and should be easily visible in the reports.
For yet other cases, where axes are in a different order, there are two cases:
- Reorienting to RAS+ produces an image of the same shape. For example, ARS+ with shape (128, 128, 64) would reorient to RAS+ with (128, 128, 64). In this case the bug should be visible but not produce a crash.
- Reorienting to RAS+ produces an image with a different shape. For example, AIL+ with shape (128, 128, 64) reorients to RAS+ with (64, 128, 128). This would actually cause a crash.
But I would like to stress that this bug only affects phase difference fieldmaps. Other files in non-RAS+ orientation are not known to cause problems.