Eyelink data

[costantinoai]

Hi,

First of all, thanks for the awesome tool.

In my group we are trying to set up some automatic workflows to reduce the manual work required to go from source data to BIDS pre-processed data, and I am looking into BIDScoin for the first part of this workflow.

In our fMRI pipeline, we often collect Eyelink eye-tracking data, as well as physiological (HR, respiration) measurements, and behavioral responses are recorded using PsychToolbox. It would be nice to set up BIDScoin in a way to automatically handle this data, and I think it would be feasible -- since the structure of the output files is predictable.

I am prepared to do some work for it, but can anyone point me to the right direction to get started?

Thanks in advance!

Andrea

[marcelzwiers]

Hi Andrea, thanks, I'm happy to hear that you are enjoying bidscoin.

BIDScoin can run both in automatic and/or interactive mode, but the interactive mode requires more advanced understanding of how things work. At my institute, not everyone uses the standard acquisition protocol, so the general workflow is to run the interactive mode once (at the start of a new study) and store the mappings in the study bidsmap. From then on, users can run biodscoiner (which is automatic) every time new subject data is collected. In principle, when you have a fixed set of acquisition protocols, you can set the mappings in the template bidsmap, and then run the bidsmapper in automatic mode, relying completely on the prior knowledge in your template (but I suppose you already know this).

I have limited experience with physiological data, although I ported this Matlab code to Python and included it in BIDScoin:
https://github.com/CMRR-C2P/MB
I once started a project to integrate phys2bids into bidscoin (ohbm/hackathon2021#12), but my phys2bids collaborator unfortunately did not contribute, so I dropped the effort altogether.

Taken together, if you are willing to make a bit of an effort and have the expertise on physiological data, then I can guide and certainly help you to write a plugin for bidscoin. The same is true for PsychToolbox, I wrote an events2bids plugin for others, but I'm not working with behavioural logging data myself.

Perhaps we should do another hackathon? :-)

[marcelzwiers]

https://bidscoin.readthedocs.io/en/latest/plugins.html#the-plugin-programming-interface

[costantinoai]

Hi Marcel,

Thanks for the info—this is very helpful!

Ideally, we’d like to run BIDScoin as a Singularity/Apptainer container on an HPC, with the job triggered automatically whenever new data is received. Most of our projects follow similar acquisition protocols, so as you suggested, it should be relatively straightforward to define the mappings in advance. Interactive mode is not our primary focus.

From a quick search, I came across this BEP 020 draft on structuring eye-tracking data in BIDS, and there’s some additional context in this PR. The discussion in the PR suggests the BIDS team is actively working on an eye-tracking converter, and that progress is being made toward a clearer structure for ET data. However, since this is still a draft, I wonder whether the current proposal would be sufficient or appropriate as a basis for developing a BIDScoin plugin?

Regarding heart rate and respiration, I found this section in the BIDS documentation. I have limited experience with these data types (Philips format), but I do have a few examples. I know that PhysIO supports these formats, which might be a good starting point to understand how to parse the files.

As for the behavioral logfiles we generate with PTB, these are custom and specific to our pipeline. So I’m guessing it wouldn’t be a priority to integrate them into BIDScoin directly. Do you have any suggestions on how best to handle them?

I’m happy to look into all this, but I’d really appreciate your input on what you think makes the most sense and how best to move forward.

[marcelzwiers]

Bidscoin is all about wrapping converters to make it easy to produce BIDS-compliant output. In other words, we need a (simple/minimal) converter before building a plugin. I'm not too familiar with physiological data, so I don't know which converter to pick. However, if you pick one, then I am happy to work with you to build a BIDScoin plugin that wraps the converter. How do you convert your eye-tracking data now?

[marcelzwiers]

p.s. with converter I mean something like dcm2niix, which converts e.g. a single input file to a single output file of a different format

[costantinoai]

The eye-tracking data we have comes from the Eyelink system. By default, Eyelink produces a proprietary *.edf file, but it can (and in our case does) also generate a human-readable *.asc file. As far as I know, there are currently no tools that fully automate the BIDS conversion of these files. However, there are Python libraries that can parse the *.asc files into pandas DataFrames. For example, eyelinkio appears to be actively maintained, and eyelink_reader is another available option.

That said, finding a tool that performs full BIDS conversion may be more challenging, as a standardized BIDS extension for eye-tracking data has not yet been fully defined.

Regarding other physiological signals like heart rate and respiration, the only tool I’ve found that supports Philips physiological data is PhysIO, which I mentioned earlier. However, it also does not perform BIDS conversion, and it requires MATLAB and it is shipped with the TAPAS.

I understand this isn’t ideal, and what we really need are converters, not just parsers. Unfortunately, I haven’t come across any ready-to-use conversion tools for these data formats. That said, I’m willing to put in the work to get us to a point where we can implement BIDS conversion for physiological data using BIDScoin, even if it means developing a minimal tool from scratch to organize the conversion process.

If this is indeed what we need, please let me know and I’ll begin working on it. Also, if you have any suggestions for how to bring in others to help with this effort, I’d appreciate it—developing a full conversion tool will take time, and working solo could quickly become a bottleneck.

And thanks for the help and clarifications provided so far!

[costantinoai]

update: scanphyslog2bids seems to handle respiration and heart-rate data from Philips scanners. Other tools to handle data from other vendors data can be found in this list. For eye-tracking *.edf files, bidsphysio.edf2bids seems to be a solution, but I did not test it.

[marcelzwiers]

I'm happy to help wrap the tool into a plugin, but I don't know the tool myself. I think it is best if you would pick the best converter for your use case, then we can build a plugin together?