Improve image intensity homogeneity using SPM
Guidelines by
S. N. Sotiropoulos
SPM
uses a Bayesian MAP (maximum a-posteriori) approach to correct for the field
inhomogeneities. The inhomogeneity correction is part of the tissue segmentation
process that is implemented in SPM5. For details on this unified segmentation
approach, see
(Ashburner J & Friston KJ, NeuroImage 26:839-851, 2005).
Since the approach is Bayesian, template tissue segmentation masks are used to
provide prior information. Since these templates provide a starting point for
the algorithm, the more our image matches these templates, the more accurate the
solution will be. Therefore, we need to make sure that the image to be corrected
has the same orientation as the templates and origin close to the origin of the
templates.
- Launch SPM.
- Make sure that the original image matches the template orientation,
i.e., xy plane is axial, xz is coronal and yz is sagittal. To check the
orientation of a template and where its origin is, we can use the "Display" tool
of SPM.
-
We can further check whether our image and the template are similar in terms of
orientation and origin, by using the "Check Reg" tool of SPM. When we launch
this tool, we open our image and the template (located in /SPM/templates/). We
should then be able to see both of them having the same orientation, in the
windows that appear. If one of the two images is not visible, then there is
probably a problem with the origin of our image. Both origin and orientation can
be changed either using
MRIcro,
or the "Display" tool of SPM.
-
We can then move on with segmentation/inhomogeneity correction. Launch the
segmentation toolbox by choosing "Segment" in the Spatial pre-processing window.
-
In the new window, choose "Data" (on the left subwindow) and then press "Specify
files" (on the right subwindow). Choose as input the image to be corrected.
-
Choose then "Output files". The options (GM/WM/CSF) indicate whether respective
tissue masks will be saved as output and in which space (native, normalised,
etc.). To choose the space, press "Specify Menu Item" on the right
subwindow. The option "Bias Corrected" indicates whether an inhomogeneity
corrected image will be saved. Again, press "Specify Menu Item" on the right
subwindow to choose "Save Bias Corrected".
-
Choose then "Custom". Here we specify parameters for the Bayesian model. For
inhomogeneity correction, we are interested in two parameters:
"Bias regularisation" and "Bias FWHM".
The idea is that the inhomogeneity is modelled as an extra field (bias field)
that modulates the true image intensity. These two parameters control how smooth
the bias field is. At lower magnetic field strengths (3T), the bias field varies
smoothly across the image, so we need light-medium regularisation. However, for
higher magnetic field strengths (7T), the bias field varies very rapidly, so we
need almost no regularisation.
To summarize:
- Very high inhomogeneity => Choose "extremely light/no regularisation" and a
small value for FWHM.
- Very low inhomogeneity => Choose "medium/heavy regularisation" and a large
value for FWHM.