Nested DICOM sequence values not being included in BIDS

[captainnova]

Hi, we have lately been receiving enhanced DICOM from a Siemens XA60 site where the locations of some fields have moved from their customary top level addresses to inside nested sequences. Unfortunately dcm2niix is not noticing them there, and they are missing from the produced BIDS files.

Example:

(5200, 9229)  Shared Functional Groups Sequence
   (0018, 9112)  MR Timing and Related Parameters Sequence   1 item(s) ----
      (0018, 0080) Repetition Time                     DS: '4800.0'
      (0018, 0091) Echo Train Length                   IS: '243'
      (0018, 1314) Flip Angle                          DS: '120.0'
         ---------
...
  (0018, 9115)  MR Modifier Sequence   1 item(s) ---- 
...
     (0018, 9079) Inversion Times                     FD: 1550.0
      (0018, 9081) Partial Fourier                     CS: 'YES'

If I unenhance with dcuncat, everything from (5200, 9229) gets moved out of it, and dcm2niix puts them in the BIDS, so I can make a diff and see the "missing" fields:
ImagingFrequency
ScanningSequence
SequenceVariant
ScanOptions
ImageComments
TablePosition
RepetitionTime
MTState
SpoilingState
InversionTime
FlipAngle
CoilString
PercentPhaseFOV
PercentSampling
EchoTrainLength
PartialFourierDirection
PhaseEncodingSteps
FrequencyEncodingSteps
PhaseEncodingStepsOutOfPlane
ParallelReductionFactorInPlane
ParallelAcquisitionTechnique
PixelBandwidth
InPlanePhaseEncodingDirectionDICOM

This is not happening for all Siemens XA60 sites sending enhanced DICOM, and we do not know why this site is different. However, the behavior of moving something to inside (5200, 9229) Shared Functional Groups Sequence seems like something much more likely for a scanner manufacturer to do than a PACS or other downstream modifier, which tend to not write in enhanced format. I noticed issue:380, but this seems to be new behavior for Siemens XA. Unfortunately the X Protocol is absent, so I cannot find the software subversion, but I suspect it changed in September.

I would like dcm2niix to also check for kInversionTimes, etc. inside nested sequences (or at least (5200, 9229) ones) so that we would not have to unenhance. I see where the "d" structure is updated around nii_dicom.cpp:6122, but do you have any advice on generalizing it to include nested sequences?

dcm2niix versions:
v1.0.20251017 (JP2:OpenJPEG) (JP-LS:CharLS) GCC8.5.0 x86-64 (64-bit Linux) (git pulled from development a few minutes ago)
v1.0.20240327

Thanks,

Rob

[captainnova]

dcm2niix used to find the BIDS fields in nested sequences, and the problem started between 20230609 and 20240325. Here's a diff of the BIDS .jsons from those dcm2niix versions for a recent Siemens XA FLAIR in enhanced format:

4d3
<   "ImagingFrequency": 123.24993,
9c8
<   "BodyPartExamined": "BRAIN",
---
>   "BodyPart": "BRAIN",
12a12
>   "StudyDescription": "122021x2025119x3TxCLARiTI",
15,17d14
<   "ScanningSequence": "SE\\IR",
<   "SequenceVariant": "SK\\SP\\MP",
<   "ScanOptions": "IR\\PFP\\FS",
26,30d22
<   "RepetitionTime": 5,
<   "MTState": false,
<   "SpoilingState": true,
<   "InversionTime": 1.8,
<   "FlipAngle": 120,
32,62d23
<   "PartialFourier": 1,
<   "BaseResolution": 256,
<   "ShimSetting": [
<       1782,
<       -11196,
<       7976,
<       31,
<       96,
<       -512,
<       36,
<       29  ],
<   "TxRefAmp": 271.195,
<   "PhaseResolution": 1,
<   "ReceiveCoilName": "HeadNeck_64",
<   "ReceiveCoilActiveElements": "HC1-7;NC1,2",
<   "CoilString": "HeadNeck_64",
<   "PulseSequenceDetails": "%SiemensSeq%\\space",
<   "RefLinesPE": 28,
<   "CoilCombinationMethod": "Adaptive Combine",
<   "MatrixCoilMode": "GRAPPA",
<   "PercentPhaseFOV": 93.75,
<   "PercentSampling": 100,
<   "EchoTrainLength": 271,
<   "PartialFourierDirection": "COMBINATION",
<   "PhaseEncodingSteps": 240,
<   "FrequencyEncodingSteps": 256,
<   "PhaseEncodingStepsOutOfPlane": 182,
<   "AcquisitionMatrixPE": 240,
<   "ReconMatrixPE": 240,
<   "ParallelReductionFactorInPlane": 2,
<   "ParallelAcquisitionTechnique": "GRAPPA",
64d24
<   "PixelBandwidth": 781,
66d25
<   "PhaseEncodingDirection": "i",
74d32
<   "InPlanePhaseEncodingDirectionDICOM": "ROW",
76c34
<   "ConversionSoftwareVersion": "v1.0.20230609"
---
>   "ConversionSoftwareVersion": "v1.0.20240325"

Obviously I wish I had included DICOM like this in my test suite back in early 2024, but it feels like we did not start experiencing it until 2025; we may not have had DICOM like it in 2024, but there is a flood of it now.

20230609 and 20240325 are just the closest old versions we had lying around; I did not do an exhaustive bisection.

[jawaltz]

Hi-

I am still getting the error below, when I try to run dcm2niix (most recent version) on DICOM files written by Siemens XA60 software.

dcm2niix is getting called by bidskit.

Any thoughts about what I can do? Is it easy to "unenhance" all the DICOM headers in a dataset?

Thanks,

Jim Waltz

Traceback (most recent call last):
File "/usr/local/bin/bidskit", line 8, in
sys.exit(main())
^^^^^^
File "/usr/local/lib/python3.12/site-packages/bidskit/main.py", line 356, in main
d2n.organize_series(
File "/usr/local/lib/python3.12/site-packages/bidskit/dcm2niix.py", line 209, in organize_series
tr.purpose_handling(src_meta,
File "/usr/local/lib/python3.12/site-packages/bidskit/translate.py", line 106, in purpose_handling
scan_seq = bids_meta['ScanningSequence']
~~~~~~~~~^^^^^^^^^^^^^^^^^^^^
KeyError: 'ScanningSequence'

[captainnova]

Hi @jawaltz , the short answer is yes, and there is a good discussion of enhanced to "classic" DICOM conversion here: pydicom/pydicom#1705

However,

• how or whether to move tags around when going from enhanced to classic isn't fully specified. XA60 enhanced -> dcuncat -> classic does not produce the same results as writing the DICOM in classic/compatibility mode on the scanner.
• classic will (likely) take more disk space.
• dcm2niix used to find BIDS values in nested sequences, so I am hoping that functionality can be restored soon.

[jawaltz]

Hi @captainnova -

Does this mean that the error I am currently experiencing is common and appropriate?

dcm2niix used to find BIDS values in nested sequences, so I am hoping that functionality can be restored soon.

I have never used dcuncat, but is this a function one can run BEFORE dcm2niix/bidskit, to flatten all the DICOM headers in a folder?

Thanks,

Jim Waltz

[neurolabusc]

@jawaltz and @captainnova you can either provide a PR or provide a DICOM that provides an explicit example of this behavior. (feel free to share with my institutional email. I do think some of the changes were made intentionally to protect against edge cases that exist in the wild. Unfortunately, usage and creation of these nested attributes is used differently by different tools. It would be great to document these uses and try to define a way to autodetect. In the short term, we may need a flag that determines whether or not these values should be used or given precendence.

[captainnova]

Hi @neurolabusc, I sent your institutional email a download link through Leapfile, (as required by my employer).

Thank you for your attention - I would be happy to submit a PR, but could use a hint at where in the code I should probably look, and advice on what the edge cases you mentioned might be. I does seem to be more complicated than I what I initially thought (not looking in nested sequences).