[Camino-users] camino with HCP data

Mon Dec 9 02:01:53 PST 2013

Hi,

We'll take a look into making the necessary adjustments to Camino to handle the spatially varying b-value; probably a bit fiddly, but doable.  In the meantime, I think you can get a reasonable feel for how well the different processing techniques will work by just ignoring the spatial variation - not a long term solution though.

I see no reason why shredder won't work with longs.  Its three arguments are just in bytes, so can't you just use a <chunksize> of 8?

Danny

On 29 Nov 2013, at 13:58, E.H.J. Nijhuis <emil.nijhuis at gmail.com<mailto:emil.nijhuis at gmail.com>> wrote:

Dear Danny and Philip,

thank you very much for your quick and very helpful answer.

I am aiming to produce streamlines which I can use for a whole brain connectivity analysis for the publicly available HCP data (multi shell data with b-values of about 0, 1000, 2000 and 3000). With the dataset it was possible to estimate FA values in camino but I ran into difficulties when processing the data with the PAS approach.

The first thing which was problematic when processing a HCP dataset was that the shredder function is not able to accept longs (raw hcp.bfloats are rather big >4GB). There was no easy and clean fix to the issue because the function uses DataInputStream.read() which cannot accept longs either. I overcame the issue by downcasting longs to int somewhere posthoc in the source, but that is certainly not robust enough for an actual fix.

The error considering the lag of b=0 images was with the mesd function which I called with the following options:

mesd -schemefile hcp.scheme -filter PAS 1.4 -fastmesd -mepointset 16 -bgmask hcp.Bchar -maskdatatype ubyte < hcp_SliceX.Bfloat > hcp_SliceX.Bdouble

This resulted in the warning "WARNING: No b=0 data, cannot normalize", while the output was essentially blank. I guess if I set the lowest b-values to 0 it will work, but given your elaborate answer I now feel reluctant to process the data any further because of the caveats with the spatially varying b-value and gradient directions.

Which brings me to my next set of questions:

1. Are you planning to bring out an update which will be able to process the HCP dataset (or more generally datasets with spatially varying gradients)?
2. If the goal is to produce streamlines for a full brain connectivity analysis: what would be your preferred way of processing for a multi-shell dataset? (I know it is rather impossible to answer this question properly, but I would be very interested in your and others users opinion and advice).

Thank you a lot for your consideration and expertise.

With kind regards,

 Emil

On Mon, Nov 25, 2013 at 5:28 PM, Alexander, Daniel <d.alexander at ucl.ac.uk<mailto:d.alexander at ucl.ac.uk>> wrote:
What are you trying to do?  Most model fitting routines should work fine without a b=0, but I guess some of the other reconstruction algorithms may demand a b=0 image.  Assuming the low b-value images in the HCP data are b=0, ie set the entries in the schemefile to zero, I would expect to give very similar results to accounting for the slightly non-zero b.

When you say "DSI acquisition", do you really mean DSI, ie a grid of locations in q-space?  HCP use a three-shell HARDI protocol, which is quite different.  Or do you have some special data?

I don't know if anyone has tried processing the standard HCP data using Camino.  Most Camino algorithms should run fine on data acquired with their protocol, although currently Camino doesn't account for the spatially varying b-value and gradient directions in the HCP data.  I am not sure how much of an issue that is in terms of results, but you should at least be able to get an output to look at.

Danny

PS: Please write to camino-users at www.nitrc.org<mailto:camino-users at www.nitrc.org> for Camino questions rather than direct to me.  You will get faster and better answers!

On 25 Nov 2013, at 10:06, "E.H.J. Nijhuis" <emil.nijhuis at gmail.com<mailto:emil.nijhuis at gmail.com>>
 wrote:

Dear Daniel,

I recently have attempted to process the diffusion dataset of the human connectome project.

Unfortunately I noticed that camino was unable to process this dataset due to the lag of actual b=0 values (see attached the sample fsl and corresponding camino tables). The small b-value variations are due certain hardware  aspects (see Sotiropoulos et al., Neuroimage 2013).

The lag of actual b0 values results in the following error messages during reconstruction:
WARNING: No b=0 data, cannot normalize

Could you please advice me how to best handle the HCP data? Simply round the 'b0'-values? Also, since it is an DSI aquisition protocol, would you still recommend me to use camino?

Thank you a lot for your expertise and consideration.

Yours sincerely,

 Emil

--
----------------------------------------------
Postdoctoral Researcher - MR Techniques in Brain Function
Donders Institute: Centre for Cognitive Neuroimaging

tel  +31 (0)24 36 10885<tel:%2B31%20%280%2924%2036%2010885>
fax +31 (0)24 36 10989<tel:%2B31%20%280%2924%2036%2010989>
web http://www.ru.nl/donders/

Postal address:  Trigon 204, P.O. Box 9101, 6500HB Nijmegen, The Netherlands
Visiting address: Trigon, Kapittelweg 29, 6525EN Nijmegen, The Netherlands
<bvals><bvecs><hcp100408.scheme>

--
----------------------------------------------
Postdoctoral Researcher - MR Techniques in Brain Function
Donders Institute: Centre for Cognitive Neuroimaging

tel  +31 (0)24 36 10885<tel:%2B31%20%280%2924%2036%2010885>
fax +31 (0)24 36 10989<tel:%2B31%20%280%2924%2036%2010989>
web http://www.ru.nl/donders/

Postal address:  Trigon 204, P.O. Box 9101, 6500HB Nijmegen, The Netherlands
Visiting address: Trigon, Kapittelweg 29, 6525EN Nijmegen, The Netherlands

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.nitrc.org/pipermail/camino-users/attachments/20131209/1ac4270a/attachment.html>