It sounds like there is confusion about modulation. Modulating T1 images
prior to segmenting makes no sense, and degrades the segmentation performance
as it needs to do more bias correction. The segmented normalised data
themselves should be modulated instead.
If you ignore the bias correction part, and the use of tissue probability
maps, then segmenting can be thought of thresholding according to the
intensities of the data. Recaling (modulating) followed by thresholding is
not the same as thresholding followed by rescaling.
Best regards,
-John
> We would highly appreciate your comment/advice on the following VBM
> questions.
> We compared GM segmentations between patients with depression and controls
> using the Christian Gaser cg_create_template and cg_vbm_optimised VBM
> scripts. We used the final GM outputs additionally modulated by the
> Jacobian determinants of the deformation fields.
> In our first analyses creation of study specific templates and GM
> segmentations were computed with normalisation option set to "preserve
> total". The final results indicated a clear reduction of GM volume in the
> right parahippocampal area for the patients.
> In a subsequent analysis we repeated cg_create_template and
> cg_vbm_optimised, but now with "preserve concentration", the common choice
> for VBM. Unfortunately, however, the interesting hippocampal volume effect
> almost completely vanished.
> To our understanding what happens if we run the cg scripts with "preserve
> total":
> - In cg_create_template the normalised T1 MRIs of each individual are
> modulated based on the GM deformation parameters. The modulated T1s are
> then segmented using the standard .mnc priors. After averaging across
> individuals a modulated T1 template and GM, WM and CSF priors are obtained.
> The priors are probability maps, so in essence they do not contain
> modulations themselves. However, since they are derived from the modulated
> T1's they are probably slightly different than with normalisation set to
> "preserve concentration".
> -In cg_vbm_optimised the T1 MRIs in native space are normalised and
> modulated based on the GM transformation parameters. Then the modulated
> images in MNI space are segmented, using our custom, modulated, T1 template
> and GM, WM, CSF priors. Again the resultant segmentations will in fact not
> be modulated, but probably somewhat different compared to the situation
> when using "preserve concentration". Finally we select the GM and
> explicitly modulate it by the Jacobian matrix field. It is this modulated
> GM we finally analyses.
> So, we guess that by selecting "preserve total" and doing an additional
> Jacobian modulation. We actually do not perform a modulation twice.
> However, the GM segmentation results may differ somewhat when using either
> "preserve total" or "preserve concentration". Taking it one step further.
> The hippocampal volume effect is clearly present only in the explicitly
> modulated but not in the unmodulated output from cg_vbm_optimized with
> "preserve total". Therefore it may be argued that the volume effect is
> encoded in the GM deformation fields. This field is computed by first
> aligning the GM prior to the original MRIs in native space, then segmenting
> the GM and finally compute the normalisation from GM in native space to the
> GM prior. Therefore it seems that the different results from the cg scripts
> with "preserve total" or "preserve concentration" may be primarily due to a
> difference in the GM prior (though we did not test this yet).
> With regard to this we have the following questions:
> -What could explain the effect to vanish when "preserve concentration"
> instead of "preserve total" is used? Is it because of the difference in the
> segmentations (priors), only?
> - By selecting "preserve total" we may have introduced a spurious result.
> On the other hand, looking at the bright side of life, we may have enhanced
> the sensitivity of our VBM analyses. Could there be any arguments to
> support the latter?
> - Is our initial analysis running the cg scripts with "preserve total"
> still scientifically valid?
|