ENH: Add SliceTimingCorrected and StartTime fields to BOLD metadata

[effigies]

Changes proposed in this pull request

Closes #2539.
Addresses #2477.
Implements provisional bids-standard/bids-specification#836

This does not update the toffset field in NIfTI at present.

One thing to address: when we're calculating the acquisition time (TA), we use st[-1] + (st[1] - st[0]), which treats each slice as having a duration. When we're calculating tzero, we use (st[-1] - st[0]) * frac + st[0], which treats each slice as an instantaneous acquisition. So tzero is a half "slice duration" less than half TA (tzero = (t[0] + TA) / 2 - (t[1] - t[0]) / 2).

Our tzero calculation is consistent with AFNI's default, but while we're here, maybe we want to adjust it to be tzero = (t[0] + TA) / 2. Or is the calculation of TA incorrect?

cc @MartinHebert @theoschaefer @nitschalex @oliver-contier @kellyhennigan

[theoschaefer]

Hi Chris,

I just remember that SPM defines TA as TA = TR - (TR / number_of_slices) (e.g. see here). That would result in a shorter TA (minus one slice duration) compared to your calculation. I do not know if treating slices as durations or instances is better (although intuitively durations makes more sense to me), but I would say consistency is most important. Thus, I would recommend to change either the tzero calculation to tzero = (st[0] + TA) / 2 as you suggested or treat slices as instantaneous acquisitions and keep the old tzero calculation.

[oesteban]

Just a reminder of the option we talked about in a meeting some days ago. Once that idea is set back, this LGTM.

[oesteban]

Suggested change

	{'VolumeTiming': [0.0, 1.0, 2.0, 5.0, 6.0, 7.0], 'SliceTimingCorrected': True,
	'AcquisitionDuration': 1.0, 'StartTime': 0.4}
	{'VolumeTiming': [0.0, 1.0, 2.0, 5.0, 6.0, 7.0], 'SliceTiming'=[0.4, 0.4, 0.4, 0.4, 0.4],
	'AcquisitionDuration': 1.0}

What are your thoughts about making it explicit that the SliceTiming metadata is a list of uniform values? This would be readily supported by BIDS and leaves no room for confusion - even if you try to run STC over again on these data, the algorithm should notice you are sampling on-grid all the time.

[effigies]

I opened an issue on BIDS: bids-standard/bids-specification#879