Hi all,
We are using a multi-echo sequence on a Philips MR7700 (previously Ingenia Elition X?). Our scan is fairly long, resulting in about 25GB of data and 400k files.
We are experiencing issues exporting the data out of Philips’ system in a reliable manner. For one thing, our tech informs us that the Philips scanner hard drive fills up after only 2 of our subjects. This gives us a really short window to check that data transfers appear complete (sometimes as short as 6 hours). And we’ve experienced missing data - sometimes individual files when exports don’t complete.
Has anyone faced issues like this before with large fMRI datasets?
Unfortunately, Philips DICOMs are packed with redundant information and devoid of crucial details (like slice timing, phase encoding polarity and total readout time). In my experience, the Philips hardware is outstanding, but the DICOM images they create are a major liability to their ability to close sales. As scientists, we tend to only touch DICOM data once, when we convert it to NIfTI. On the other hand, clinical teams retain DICOM data, and the bloat has impacts on transfer speeds, storage and inspection (e.g. viewing the meta data from an enhanced Philips DICOM using Osirix or Horos). I have seen many Philips images where most of the file size is devoted to the bloated redundant header rather than the image data.
Assuming you have a research site, the pragmatic solution is to export the data as PAR/REC format rather than DICOM. The PAR/REC files have a simple, minimal header which minimizes file sizes. However, you need to be aware that this minimalism may not retain many sequence details.
I would urge you and all Philips MR users to lobby the Philips Clinical Scientist associated with your centers to create streamlined enhanced DICOMs (Canon and Siemens XA provide excellent templates). As a research community, we need a competitive market to negotiate sales and spur innovation. I honestly believe that the Philips DICOM images hurt their competitiveness in the clinical domain (where the vast majority of sales are made).
@sandeepganji may also have insights.
2 Likes
Thanks, Chris. This is extraordinarily informative.
I dismissed the PAR/REC option early for the reasons you suggest - it’s proprietary and lacks critical sequence details. But is there a way to combine this somehow with the DICOM header? (E.g., verify that the NIFTI images are complete by transferring and converting the PAR/REC, but then filling in the NIFTI header or .json sidecar with the missing info from DICOM headers?)
We have also already had issues with some of the details provided in other DICOM fields that Philips refuses to include as standard and that you mentioned (Susceptibility Distortion Correction - effective echo-spacing and total-readout time for Philips data · Issue #5 · nipreps/sdcflows · GitHub). I’ll take your advice and lobby Philips. My (extremely patient) tech has suggested this as well.
1 Like
@stevenweisberg both the PAR/REC and current Philips DICOM omit some crucial details. Therefore, valid BIDS sidecars will require you to inject some information. I assume you will be acquiring the same sequence for multiple participants, and so most of the required information will be constant across your dataset. Perhaps you could use the inheritance principle to specify this just once for all datasets. I am not sure that the DICOMs have any session specific varying information that is missing in the BIDS. The role of your center’s Philips Clinical Scientist is to help with these issues. I would certainly work with them to help you develop complete BIDS datasets. This will help your team generate curated datasets and let the Philips team identify the attributes required for reproducible science.