The GUI frontend of DTK contains one dialog style window.
Move your mouse cursor over the image
on the right to get the details of each control.
A few things that need to be emphasized
- Gradient table should NOT contain b0 (0, 0, 0).
- Gradient correction is to convert the given gradient directions from gradient coordinate to
the actual image coordinate. Some manufacturers use
the fixed scanner magnet coordinate as their gradient coordinate. In that case, when there is oblique angle
applied in the scan, i.e., the image coordinate is different from the magnet coordinate, correction
must be applied to the gradients to ensure correct tensor calculation and fiber tracking.
IMPORTANT: From what we know, this correction should only apply to Siemens
DTI data and some Siemens HARDI/Q-Ball data. If the HARDI/Q-Ball data is from a generic DTI sequence, e.g., you are
trying to reconstruct a DTI dataset as HARDI,
you should apply correction. If the HARDI/Q-Ball data is from a dedicated Q-Ball sequence,
you should NOT apply correction to it.
- Automatically finding mask threshold may not always work. If you get an empty track file, it is very likely the threshold
was set to high. You will need to adjust it manually.
- If you get tracks that look apparently wrong, it is very likely because the image orientation
and/or voxel order was not set correctly
at the reconstruction step. It may also because different manufacturers store image information differently (We tested the program mainly with
Siemens data). In such case, you would need to apply orientation patch by "flipping around" the
vectors to get the tracks right.
From our limited experience, Siemens data usually do not need any orientation patches, while GE and Philips data often
need to apply "Invert Y".
Orientation patch selections may vary between different imaging models. For example, because DSI reconstruction
uses pre-calculated matrices based on Axial data, it will be different with other data.
In DSI case, the only difference is that user can only pick a pre-defined gradient table and use
a pre-calculated reconstruction matrix. At this point, we provide two gradient tables, one is 257 (plus 1 b0), the other is 514
(plus 1 b0).