Hi forum,

I have tried using DTK to analyze the VIS 2010 contest data (http://viscontest.sdsc.edu/2010/).

Unfortunatly I cannot reproduce the tracking results for the synthetic dataset they provide.

This is what I did:

Split the DTI raw data into one file per timestep and create a analyze header for each file.

The original data is in 16-bit unsigned in format, which is unsupported by DTK. Therefore I converted the volumes to float but was unsure about the data range and tried normalizing it.

I have tried various invert/swap combinations without success.

The results don´t look totally wrong – the helix structure can be recognized, but the tracks are mostly very short and the directions don´t look right.

Any ideas or hints?

Hi,

thanks for the quick reply.

The original data is just one raw file with b0 and all time steps.

I split it up into seperate raw files, named "dti_000.img" etc. and generated a analyze header for each using camino´s (http://web4.cs.ucl.ac.uk/resea…..pmwiki.php) analyzeheader command:

./analyzeheader -datadims 64 64 64  -datatype float > /data/visContestData_raw/syntheticDTIData/dti_000.hdr

Here is a screenshot:

http://img156.imageshack.us/im…..4/dtkt.png

This is the reference image:

http://img156.imageshack.us/im…..0/refb.png

This is the log:

Steps to do:

1. dti_recon "/data/visContestData_raw/syntheticDTIData/dti_000.img" "/data/visContestData_raw/syntheticDTIData/dti" -gm "/tmp/dtk_tmp/matrices/gradient.txt" -b 1000 -b0 auto -iop 1 0 0 0 0 -1  -p 3 -sn 0 -ot nii

2. dti_tracker "/data/visContestData_raw/syntheticDTIData/dti" "/data/visContestData_raw/syntheticDTIData/dti.trk" -at 1   -m "/data/visContestData_raw/syntheticDTIData/dti_dwi.nii" -0.98 1000 -rk2   -it nii

3. trackvis "/data/visContestData_raw/syntheticDTIData/dti.trk"

——— Launching dti_recon —————–

Did not find image info file '/data/visContestData_raw/syntheticDTIData/dti_.info'. Default info being used.

Volume dimension: 64 64 64
Voxel size: 1.000 1.000 1.000
Number of b0's: 1
Number of gradient directions: 30

Reading raw data… 100%
Reconstructing… 100%

b0 threshold used to mask background: -22558.2 22816

Writing output…
Done.
Reconstruction completed successfully.

Time taken: 00:00:01

——— Launching dti_tracker —————–

Volume dimension: 64 64 64
Voxel size: 1.000 1.000 1.000

Threshold for mask one:  -0.98 1000

Reading tensor data…done.

Tracking… 100%
Number of tracks: 209158
Time taken: 00:00:09

——— Launching trackvis —————–

TrackVis successfully launched.

Total time taken: 00:00:10

 

When using ushort as datatype I get the following output:

Steps to do:

1. dti_recon "/data/visContestData_raw/syntheticDTIData/dti_" "/data/visContestData_raw/syntheticDTIData/dti" -gm "/tmp/dtk_tmp/matrices/gradient.txt" -b 1000 -b0 auto -iop 1 0 0 0 0 -1  -p 3 -sn 0 -ot nii

2. dti_tracker "/data/visContestData_raw/syntheticDTIData/dti" "/data/visContestData_raw/syntheticDTIData/dti.trk" -at 1   -m "/data/visContestData_raw/syntheticDTIData/dti_dwi.nii" -0.98 1000 -rk2   -it nii

3. trackvis "/data/visContestData_raw/syntheticDTIData/dti.trk"

——— Launching dti_recon —————–
** ERROR: nifti_convert_nhdr2nim: bad datatype ** ERROR (nifti_image_read): cannot create nifti image from header '/data/visContestData_raw/syntheticDTIData/dti_000.hdr' Can not load image data from /data/visContestData_raw/syntheticDTIData/dti_000.hdr
Process aborted due to error.

Using short as datatype produces roughly the same results as using floats.

A higher angle threshold only resulted in longer lines in wrong directions :)

Hi,

the endianness was the problem, the tracking works fine now.

Thanks for your help and for providing dtk!