Temporal shifts of time series (BOLD, ASL, PET, *EG?)

[effigies]

An issue that has come up in fMRIPrep (nipreps/fmriprep#2477) is that slice timing correction can modify the effective time of each volume, which can impact analysis when the repetition time is a reasonably large fraction of the HRF delay. We need a way to indicate the effective time via an offset.

The NIfTI header has a toffset field, and CIFTI-2 has a SeriesStart attribute for time series axes. GIFTI time series have no such internal metadata.

Because of this inconsistency, and following the principle that critical metadata should be in the BIDS sidecar, it makes sense to allow for temporal offsets to be specified in BIDS metadata.

I would suggest that this not be considered a derivatives-only piece of metadata, as it corresponds closely to the TimeZero and ScanStart concepts introduced in PET (PET Metadata - Time). I would suggest they be generalized.

At present, I think only ScanStart would be needed, and it could be REQUIRED to correspond to toffset in NIfTI files. Is there a use case for generalizing TimeZero as well?

Here's a basic proposal:

Key Name	Requirement level	Data type	Description
ScanStart	OPTIONAL	number	Time of first sample, in seconds, relative to the start of the experiment. The default value is 0. In NIfTI files, this value MUST be consistent with the toffset header field (after accounting for units stored in xyzt_units field).

I don't know if this should apply to EEG/MEG/iEEG.

[CPernet]

in MEEG you will start recording and then start the experiment - so the 1st data sample is always before the 1st experimental trial

[sappelhoff]

(Might be interesting for @sjeung or @JuliusWelzel ?)

[effigies]

One thing that needs consideration is how this interacts with VolumeTiming. I would propose that VolumeTiming is mutually exclusive with StartTime, so that you can either define your volume onsets as VolumeTiming or StartTime + RepetitionTime * [0..NumberOfVolumes].