Difference in ReconMatrixPE between dcm2niix versions v1.0.20181125 GCC4.8.5 and v1.0.20220720

We are changing the dcm2niix version used to the latest. We had been using v1.0.20181125 for a while. I noticed quite a few differences in the metadata, including a lot of new information being saved. One thing caught my eye – when I noticed that the older version reported a different value for ReconMatrixPE (320 as opposed 280 in the latest version) for the same data. Is this something known that was a bug before ?

This question is underspecified. The reason that scientists prefer NIfTI/BIDS is it is simpler and use common terminology and format regardless of vendor. It is unclear if your data comes from Siemens, GE, UIH, Canon, or Mediso. Attaching the BIDS JSON text into your message would help resolve your issue.

In general, you can determine the source of BIDS tags here, but ReconMatrixPE is not a formal BIDS field.

You can get further details by looking at the manufacturer specific pages. I am going to go out on a limb, and guess that your data comes from a GE scanner, in particular the image interpolation settings:

  • However, dcm2niix should correctly detect this interpolation, resolving apparent discrepancies between tags (0020,1002; 0021,104F; 0054,0081). Issue 355 provides details.

For future acquisitions, it is typically better to acquire images without interpolation. Image processing will be faster, offline interpolation can be more sophisticated, and interpolation can prevent some forms of artifact removal.

I apologize for the underspecified question, @Chris_Rorden . The system is a Prisma_fit. Please see attached the two json files (suffix changed to .txt). Thanks for your help!
2022.txt (1.8 KB)
2018.txt (1.6 KB)

2018 always evaluates as true:

if  (h->dim[2] == h->dim[2])

2022 evaluates as true only if columns and rows are identical:

if (h->dim[1] == h->dim[2])

OK. So this was a bug, then, in the old code!