Hi Kristian,
Sounds like you’ve really had a thorough look into this.
Regarding the formula; we also came across an old file from philips in which there was mention of no dividing by the sense factor, so we tested with and without (.3 vs .8 effective echo spacing in our case). The unwarping results did not look OK when we did not divide by the sense factor, the registration looked like the compensation for the distortion was too much (I remember thinking the brain looked ‘squeezed’ in a sagittal view).
We guess that all we can do is recommend people to run comparative tests like this until we know whether the Philips field maps are standardised or not.
Good luck!
Anna & Anne
Dr. Anna van 't Veer
\ Child and Family Studies
\ Leiden Institute for Brain and Cognition (LIBC)
\ Social and Behavioural Sciences
\ Leiden University
On 04 Aug 2017, at 18:58, Kristian Loewe <[log in to unmask]> wrote:
Hi,
I am also currently analyzing some Philips 3T data for the first time.
The manual by Anna van 't Meer et al. was very useful to me. Thanks, Anna!
I have a few remarks and questions, though (see below).
1) In my case, I received the data in DICOM format (not PAR/REC) from a collaborator and converted them to NIFTI using dcm2niix (not dcm2nii), which also creates descriptive json-files in addition to the data in NIFTI format.
One can use the json-files corresponding to each image to figure out automatically which of the images is the phase and which is the magnitude image (which could maybe supplement Anna's manual, section 4) using
$ cat $jsonfile | grep PhilipsRescaleIntercept \
| awk '{print substr($2, 1, length($2)-1)}'
The phase image is the one for which the above command yields a non-zero rescale intercept (in my case it's -217).
2) It seems that the Philips fieldmap comes as a phase difference image in Hz ranging from -0.5/deltaTE to +0.5/deltaTE. Thus, I converted it to radians using
fslmaths phasediff_orig.nii.gz \
-mul 2 -mul 3.14159 -mul 0.0023 phasediff_rad.nii.gz
(using deltaTE = 0.0023). The result ranges from -pi to pi. See https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=fsl;5a9ff5d5.1310.
In Anna's manual, the multiplication with deltaTE is not applied.
3) Regarding the effective echo spacing, I'm not sure whether the formula in Anna's manual (page 7) is correct. I got an old Philips manual (from 2010) from the local MR team, which states
"Since the ETL takes care of the SENSE factor, you don't
need to do any further calculation to calculate an
effective dwell time"
As a result, it seems to me that you don't have to divide by the SENSE factor. In my case, I computed the effective echo spacing as
ES = (1000 * 12.499) / (434.2756 * (39 + 1)) = 0.7195
4) When analyzing Siemens data, I often used fsl_prepare_fieldmap to prepare the fieldmap for subsequent use with epi_reg. I was thus wondering if the FSL team would still be interested in a collaboration to extend the functionality of fsl_prepare_fieldmap to be able to handle Philips data too (as is mentioned on the wiki, https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FUGUE/Guide#Non-SIEMENS_data).
Looking at the code in fsl_prepare_fieldmap, it seems to me that one could basically use the existing Siemens routine from the "unwarp phasemap" step onward after converting the Philips phase difference image to radians.
I decided to take a closer look at what exactly happens with Siemens data when using fsl_prepare_fieldmap with one of my data sets. After dicom conversion, my Siemens phase difference image ranges from -4096 to 4092. This leads to the image being divided by 2 as part of the sanity check in fsl_prepare_fieldmap. To convert it to radians, the map is then divided by 2048, next 1 is subtracted before it is multiplied by pi.
The resulting map ranges from -2pi to 0. I wonder if this is the intended range at this point. I suspect that you expect a positive input range and subtract 1 to arrive at an output range of -pi to pi. I don't suppose it makes much of a difference for the final result anyway.
Any additional input on this would be much appreciated!
Cheers,
Kristian
On 04.08.2017 17:18, Ian Malone wrote:
> Hello Anna,
> Thank you, I'll give it a read.
> Regards,
> Ian
> On 04/08/17 15:58, Anna wrote:
>> Dear all,
>>
>> Hopefully this manual for B0 unwarping with data from a philips scanner can be of use to you: https://osf.io/hks7x/
>>
>> It took us quite some time to figure it out, but the results after correcting look good (we also ran it by Mark Jenkinson), so I'm happy that others can now use this and maybe even improve it. Please let me know if you have any questions or suggestions.
>>
>> For the specific question of shifting voxels in the wrong direction; this may have to do with the scan you acquired, the last part of the manual briefly touches on that.
>>
>> Cheers,
>>
>> Anna van 't Veer
>>
>>
>>
>> Op 4 aug. 2017 om 16:02 heeft Ian Malone <[log in to unmask] <mailto:[log in to unmask]>> het volgende geschreven:
>>
>>> Hello,
>>>
>>> I re-discovered this old post about a S/I gradient in field maps from Philips scanners, as I'm currently looking at some Philips data which displays exactly the same issue. Did you ever manage to resolve it?
>>>
>>> Best wishes,
>>> Ian
>>>
>>> On 09/07/12 16:32, Michael Harms wrote:
>>>> My apologies for re-posting this one more time, but I'm really stumped.
>>>> I would REALLY appreciate it (and would be happy to treat you to a
>>>> beverage of your choice at the next OHBM) if someone that has
>>>> successfully acquired and applied field map correction from a Philips
>>>> Achieva could contact me with what they had to do.
>>>>
>>>> I'd be happy to then summarize whatever I learn to the broader FSL-list.
>>>>
>>>> thanks,
>>>> -MH
>>>>
>>>> On Fri, 2012-06-29 at 11:38 -0500, Michael Harms wrote:
>>>>> Hello FEAT users with Philips scanners,
>>>>>
>>>>> I would be highly appreciative if someone that has successfully applied
>>>>> fieldmapping correction (B0 unwarping) using Philips scanner data would
>>>>> contact me off the list so that I could pick your brain.
>>>>>
>>>>> We have data from a multi-site study, of which one site was a Philips
>>>>> scanner. I *thought* that I knew what needed to be done to apply
>>>>> fieldmapping correction for the data from this site, but the comparison
>>>>> in the Prestats report of "undistorted example_func to undistorted
>>>>> fieldmap" clearly shows that we don't have something right.
>>>>>
>>>>> The odd thing is that the "Unwarping shift map" from the Philips site
>>>>> data shows a spatial pattern that is completely different from what we
>>>>> observe from 4 different Siemens sites. Namely, it has a rather strong
>>>>> S/I gradient, with negative voxel shifts inferiorly, and positive voxel
>>>>> shifts superiorly. This makes me wonder if what we collected wasn't a
>>>>> proper fieldmap (the site collected a FFE with "B0 field map = yes" and
>>>>> "delta TE = 2.5 ms"), but if that's the case, I'd like to know what the
>>>>> site needs to do differently at the time of acquisition.
>>>>>
>>>>> Thanks in advance!
>>>>> -MH
>>>>>
>>>>>
