search

bids-standard / bids-specification

Brain Imaging Data Structure (BIDS) Specification
https://bids-specification.readthedocs.io/
Creative Commons Attribution 4.0 International
275 stars 157 forks source link

Should files in file collections have the same dimensions/FOVs? #869

Open tsalo opened 3 years ago

tsalo commented 3 years ago

I noticed recently that, due to what seems to be a random error, multi-echo and complex-valued fMRI runs in one of my datasets ended up with different lengths in different files. The validator didn't catch this, because I don't think there's a strict rule that files within a file collection have to have the same dimensions. I think this is a reasonable rule under most circumstances, but it could be a problem for qMRI, so I figured I should check.

Pinging @agahkarakuzu and @Gilles86.

agahkarakuzu commented 3 years ago

Thank you for reaching out @tsalo!

think this is a reasonable rule under most circumstances, but it could be a problem for qMRI, so I figured I should check.

Matching volumetric dimensions would indeed be a good sanity check for file collections. I am trying to think of cases where spatial dimensions would differ within a file collection, but it is highly unlikely.

Gilles86 commented 3 years ago

Not sure I follow. Do you mean all files in the file collection should be either 3D OR 4D? Or do you mean the size of the first 3 dimensions? Or actually only the 4th dimension?

tsalo commented 3 years ago

@Gilles86 all of the above. If one file in a file collection is 3D (or 4D), they should all be. Additionally, all dimensions should match across files. In my specific case, only the fourth dimension was different, but I think it should hold for all of them.

Gilles86 commented 3 years ago

I can't up with an example but it "feels" like a dangerous restriction to me. Who knows someone wants to do different readout schemes for different parts of the parameter sweeps or something...

agahkarakuzu commented 3 years ago

Who knows someone wants to do different readout schemes for different parts of the parameter sweeps or something...

@Gilles86 that's why my understanding was to apply this rule to x,y,z.

@tsalo can we make this "recommended" and raise warnings if dimensions mismatch?

tsalo commented 3 years ago

I can't up with an example but it "feels" like a dangerous restriction to me.

That's fair. It does sound like a less restrictive rule makes more sense for anatomical MRI. At least for fMRI, though, I can't imagine a case where this rule wouldn't apply, since in that case all file collections should come from the same scan.

@Gilles86 that's why my understanding was to apply this rule to x,y,z.

@agahkarakuzu I think I misunderstood. When employing different readout schemes within the collection, what would the effect on dimensions be?

@tsalo can we make this "recommended" and raise warnings if dimensions mismatch?

I think that's the best option. I can follow up with the validator, but I'll also need to open a PR to the specification to add a note about the new recommendation.

tsalo commented 10 months ago

I've opened https://github.com/bids-standard/bids-validator/issues/1853 about this.

agahkarakuzu commented 10 months ago

@tsalo I think warning is the best course of action here. Almost without exception, woxel-wise spatial correspondence is expected across individual images of a file collection. Then again, the keyword is "almost", there is always one or two things that make an exception.

For a majority of the cases, warning should act as a good sanity check.