Dcm2niix only converting phasediff, missing magnitude1 for ADNI Siemens GRE Field Maps (Incomplete Raw Data?)

Software Support
1

Hello NeuroStars community,

I’m working on a project to convert ADNI3 sMRI and resting-state fMRI data into BIDS format, primarily using dcm2niix within a custom Python script. I’m encountering a persistent and critical issue specifically with the field map sequences for several subjects (e.g., 036_S_4820, 036_S_4894, 037_S_0150).

Observed Problem in BIDS Output: When my script processes raw DICOM folders identified as ‘SAG GRE FIELD MAPPING’ or ‘gre_field_mapping’ for these subjects, dcm2niix (v1.0.20241211) consistently only generates a _phasediff.nii.gz file and its corresponding JSON sidecar. The crucial _magnitude1.nii.gz file is never created.

Consequently, when I run the BIDS validator, I receive the following errors/warnings:

  • [ERR] You have to define 'EchoTime1' and 'EchoTime2' for this file. (code: 15)
  • [WARN] Each _phasediff.nii[.gz] file should be associated with a _magnitude1.nii[.gz] file. (code: 92)

(Other warnings like SliceTiming and EVENTS_TSV_MISSING are present but are secondary at this point, as they are likely expected or can be addressed later.)

My Diagnostic Findings (using dcm2niix -v 2 on raw field map DICOM folder): For a problematic raw field map DICOM folder (e.g., C:\small DB\036_S_4820\SAG_GRE_FIELD_MAPPING\2012-07-17_13_26_28.0\I316731), the verbose dcm2niix output shows that all 45 DICOM files in that folder are consistently identified as: ph 1 mag 0 echo 2.

This indicates that the raw data I possess for this ‘field mapping’ acquisition contains only phase data from the second echo, and completely lacks any magnitude data from either echo.

My understanding of the root cause: It appears the raw DICOMs downloaded for these field map acquisitions from ADNI are incomplete. For dcm2niix to produce a valid BIDS field map pair (_magnitude1.nii.gz and _phasediff.nii.gz with EchoTime1/EchoTime2), it requires both magnitude and phase data, typically from two echoes. Since the input only contains phase data from a single echo, dcm2niix cannot form the complete pair. This suggests an issue with the original data acquisition or, more likely, how I’ve selected/downloaded the data.

My questions to the NeuroStars community:

  1. Is this a known characteristic or common issue for Siemens SAG GRE Field Mapping sequences (especially older ones from ADNI)? Does it sometimes produce phase and magnitude components as separate series that need to be explicitly combined?
  2. How can I reliably identify and download the complete field map series from the ADNI LONI IDA website? When looking at the ADNI IDA interface, what specific criteria (e.g., “Image Data ID” patterns, “Description” variations, or other fields) should I use to ensure I’m downloading all necessary DICOMs (magnitude and phase, both echoes) that dcm2niix can correctly process into a BIDS field map pair?
  3. If a complete field map series is genuinely unavailable for certain subjects in ADNI, what are the recommended best practices for managing this in a BIDS-compliant workflow?
  • Should I simply omit these subjects from analyses requiring field map correction?
  • Is there a way to generate synthetic field maps if the raw acquisition isn’t complete (and is this BIDS-compliant/acceptable for fMRIPrep)?
  • What’s the best way to document such missing data for BIDS validation purposes, short of removing the entire fmap directory for those subjects?

Any insights or guidance would be greatly appreciated! Thank you.

2

Hi @Alireza217,

I do not know enough about ADNI to help and without seeing your python script or dcm2niix outputs from the dicoms it will likely be hard for those not familiar with ADNI to help.

I can share thoughts about other points though. You can try to generate a synthetic fieldmap with SynBOLD-DisCO and organize it like a BIDS style reverse phase encoded epi.

I don’t know enough about your analyses to say one way or the other. I would recommend processing all subjects the same way to remove doubt about this.

You can prepare BIDS filter files for software that ignore the fmaps if you decide to do that. Or you can just move the fmap data to sourcedata. Whatever method you take, just make sure to detail it in the README.

Best,
Steven

1 Like
3

@Alireza217 the ADNI datasets have been collated from many sites, with various stages of manipulation and anonymization. In addition, a tiny subset have been corrupted in ways that make them hard to repair.

For this reason, is is suggested that you ask ADNI related questions are welcome at ADNI | Ask the Experts , where questions are forwarded to the appropriate experts according to their category.

@captainnova may have further insights.

1 Like
4

Hello @Alireza217, Prof. Rorden is basically correct that since ADNI is a multisite study, it sometimes encounters issues with local protocol quirks and DICOM transmission, which we and the sites put a lot of effort into correcting ASAP. However, the field maps were both particularly problematic and have a lower priority than actual science scans, and we eventually stopped acquiring them.

I checked 036_S_4820 20120717, and yes, all we received was a single phase (or phase difference) volume, which was a fairly common problem. At the time we were considering substituting another image for the magnitude, but now that there is SynBOLD-DisCO I think that would be a better route. Thank you for suggesting it, @Steven. I have not used it personally, but it has to be a better approach than expecting the field maps to always be there and usable. Besides the missing magnitude problem, head coverage and interscan registration were other problems.

Best wishes,

Rob

2 Likes