Data formats review

[matthew-brett]

From : February 9, 2022, 1:03pm

This topic is to organize the next in our short series of meetings on
neuroimaging data formats.

At the last meeting, we discussed the ADSF
paper,
and thought about desirable properties of a new data format.

At the end of the meeting, we agreed that it would be useful to discuss
the various data formats in common use for analysis at the moment, with
their benefits and dysbenefits. Let's say those are:

• NIfTI (Chris Rorden)
• MINC
• AFNI
• DICOM

We said it would be good to have a short talk about each, followed by
discussion.

The first question is - who would like to talk about these formats?

I guess we'll end up with a meeting in a few weeks, if the stars align.

[matthew-brett]

From: February 9, 2022, 1:31pm

For the presentations, I suggest we do something like:

• What is the format?
• Who's using it?
• What is it really good at?
• What do you miss when you use it?
• If you were making a 2.0 version of the format, what would change?

[matthew-brett]

@ neurolabusc on February 9, 2022, 2:40pm

I would be happy to chat about NIfTI. I have some draft slides on Google Drive. If anyone wishes to extend or amend my comments, please contact me and I can provide editorial permissions.

I would suggest you add DICOM to the list of formats to discuss. I would
suggest Andrey Fedorov could present this, and perhaps describe his
dcmqi (DICOM ( dcm ) for Quantitative Imaging ( qi)).

[matthew-brett]

On February 9, 2022, 2:59pm

Chris - great - thanks for volunteering for NIfTI, that's very helpful.

I did think of DICOM, but it seems to me that its currently very rare to
use that as an analysis format - do you agree?

[matthew-brett]

@ neurolabsc on February 9, 2022, 8:33pm

Matthew - I agree that DICOM analyses are very rare now. However, if our remit is
to discuss future formats, it seems an opportunity to discuss whether
the dominant format used clinically could fill this niche. One could
leverage a huge amount of existing infrastructure. It is not only a
format, but a proven method for storage, query and transfer.

It is true that DICOM is historically very complex, with many
compression strategies that are not seen elsewhere. It has been
interpreted differently by different vendors. Further, vendors have
introduced errors over the years, so robust tools need to include
kludges for these. Classic DICOM saves each 2D image as a separate file,
which can overwhelm file systems.

On the other hand, Canon and Siemens (XA) have shown how modern enhanced
DICOM can save an entire series in a lean single file. Since DICOM only
requires compliant tools to handle the uncompressed transfer syntax, one
can envision a future DICOM that is easy to support robustly.

If our remit is to envision future formats, I would advocate that we
seriously consider if cooperation with
WG-16
could benefit both clinical and research communities. From my
perspective, the last couple decades have seen a growing divergence
between the research and clinical communities. I would argue we should
seek a virtuous cycle that would allow much easier translation of
research tools to change the standard of care.

[matthew-brett]

@ pieper on February 10, 2022, 12:55pm

I share neurolabsc's perspective on this. DICOM is far from perfect and is often
frustrating, but that can be said of other formats too if we're honest.
And DICOM will always be part of the neuroimaging discussion because it
is what will come out of the scanners for the foreseeable future. I'd be
interested in leveraging things like zarr and ngff for programming
convenience and performance without losing the standardized
representation of acquisition parameters, subject ids, demographics,
etc.

[matthew-brett]

On February 10, 2022, 1:14pm

Just to clarify - I was thinking specifically of gathering experience of
using these formats in analysis - rather than reviewing potential
successor formats - which I suppose could include the new DICOM. I'd
prefer to stick to discussions of the format itself rather than
particular software to support it - although that will obviously be
tough for AFNI.

But - having said that - I can see the argument for including DICOM, in
reviewing what it is good at and what it is bad at.

So let's add it to the list - I've edited the original email above - and
I've taken the liberty of signing you up (Chris) for Nifti.