Brainshift Correction
Last updated
Last updated
Context
This part of the processing pipeline is a tool to reduce bias introduced by brain shift. Brain shift happens when the skull is opened during surgery, it means that the brain (often nonlinearly) moves with respect to the skull, e.g. due to pneumocephalus (Image 1).
If you want to coregister the whole postop CT image to a preop MRI of the same patient, linear coregistration may result in a good match of the skull but a wrong coregistration of the brain, especially in frontal regions. Most often in DBS, we are interested in subcortical regions that could be seen as "remote enough" from the pneumocephalus. However, sometimes, there is no substantial brain shift or pneumocephalus to be found and it could be okay to not correct for this issue.
One common strategy in neuroimaging would be to use nonlinear deformations instead of linear transforms. However, in DBS, this is not possible since the electrodes in the postop image would be considered part of the tissue and could be nonlinearly moved within the brain. If this is not clear to you, think about why this is exactly what we do not want: Since we are interested in the relative location of the DBS electrodes with respect to other brain structures, we should never apply a nonlinear transform between postop and preop images.
A solution that may drastically reduce the bias introduced by brainshift is to use linear transforms but apply them to subcortical regions of interest only (Image 2).
To run brainshift correction, you first go through rough coregistration and normalization to template space. Select the patient (arrow 1), tick Brainshift correction with a mask, Check Results and press Run.
To estimate a transform, choose either No Mask, Coarse Mask or Coarse + Fine Mask. No Mask will just use the cropped images to estimate the transformation and is not recommended. By choosing the Coarse Mask, a larger mask will be applied. If you choose to use both, Coarse + Fine Mask. The coarse mask (blue mask in the Schönecker publication within the #context panel) will be applied first, followed by the finer (yellow within the #context panel) mask.
In our example, let's see the use of the Coarse Mask. This will produce the following result:
As highlighted by the yellow arrows, some regions better overlap in this refined transform. The estimated transformation matrix is printed in the top right corner of the figure.
If you think the approach improved results, click Approve & Close to apply the transform to your DBS electrode reconstructions. If not, click Disapprove & Close. Alternatively, you can change the settings to No Mask or Coarse + Fine Mask and press (Re-)compute coregistration using... .
Output
A pop-up window with information about methods and references. If this information is not needed, the window can be closed.
New normalization data will appear in the selected file, under derivatives/leaddbs/patient_name/brainshift/. Anat folder contains results, checkreg folder stores the images as .png files
In theory, subcortical refines can be applied together with coarse refines "in one go". We used this strategy in earlier versions of Lead-DBS. However, the process was not robust enough and could not be implemented using all software available in Lead-DBS, in the same way. Furthermore and especially when dealing with postoperative CTs or significant electrode artifacts on MRI, many users sometimes manually apply a whole-brain coregistration in different software (such as if the options in Lead-DBS do not generate satisfactory results). That's why we chose to include this refine step at the very end of our pipeline. First ensure an as good as possible whole-brain coregistration and normalization then apply this subcortical refine step to the data.
As a side note: This processing step was completely implemented into Lead-DBS during the . Many thanks go out to the organizers of the event – as always @ brainhack, it was phenomenal.
You can use a slice viewer (such as e.g. to further examine results in detail. The relevant files will be in derivatives/leaddbs/patient folder/brainshift/anat.
