there might be different contributors that can explain the volume differences:
1. The Partial Volume Estimation (PVE) will mostly affect GM and WM volume. Now, the fraction of a pure tissue type is given in each voxel rather than the density/probability as used in older versions. Many voxels might contain a mix of two tissue types (e.g. basal ganglia), which will influence local volumes.
2. The omission of any tissue priors in VBM8 will also affect the segmentation and the resulting volumes. This effect will be more obvious for older or younger subjects, that deviate from the healthy control sample used for the creating the tissue priors in SPM.
3. The new skull-stripping approach will remove more CSF outside the brain. Thus, resulting CSF volumes are smaller.
The effects you have noticed in your attached image should be interpreted with caution because of the applied modulation. The modulation will change the local intensities in your segmentations, so that some areas (in your case the corpus callosum or frontal pole) appear darker or brighter than expected. Furthermore, the effect of modulation is also larger compared to older segmentation approaches, because the deformations used for spatial normalization have more spatial resolution and are more detailed. Thus, the extent of local volume changes (=modulation) will also increase.
To check the quality of the segmentation the un-modulated images are more appropriate.
The stripes in your images are also due to modulation and will be also visible with the New Segment toolbox + Dartel.
The two main features in VBM8 (PVE + omission of priors) will usually improve segmentation accuracy and the resulting segmentations should be better reflect the original brain anatomy. I would rather compare the segmentations with the anatomical image to check whether segmentation was successful or not.
Christian Gaser, Ph.D.
Assistant Professor of Computational Neuroscience
Department of Psychiatry
Friedrich-Schiller-University of Jena
Jahnstrasse 3, D-07743 Jena, Germany
Tel: ++49-3641-934752 Fax: ++49-3641-934755
e-mail: [log in to unmask]
On Sat, 29 May 2010 01:15:57 +0800, lion gao <[log in to unmask]> wrote:
>Dear Christian Gaser,
>I did get the raw volumes directly from the text file. Here I made a table
>of comparison of GM/WM/CSF volumes betweenVBM5 and VBM8. Attached please
>find the file for reference. Actually, one may detect by eye in the figure I
>attached previously that the GM volumes is different between VBM5 and vbm8.
>With regard to the density I mentioned in MRIcro, I asked the question since
>it looks a bit strange in WM segmentation of VBM8: the white matter near
>frontal pole have a high density(lighter), while those of corpus callosum
>seems to have a lower density(dimmer). It may be not very reasonable, as
>corpus collosum should have higher density, at least than white matter in
>frontal pole? One additional problem i just find is that the WM segmentation
>image is strip-like in VBM8. Is this normal The one in VBM5 dosen't have it,
>as it is all white. This may not be due to movement, as young subject
>usually keep quiet during scanning. The data was collected in 3T Philips
>scanner, with 1*1*1.5 mm resolution. I may need to try all the rest subjects
>to see where the differences are by chance, though it seems more like a
>With a clinical background, i am not familiar at all with those technique
>things. What I'd like to know is which one, VBM5 or VBM8 is more reliable,
>so that I may follow it. Thanks very much again for your further explanation
>On 28 May 2010 06:28, Christian Gaser <[log in to unmask]> wrote:
>> Dear Gao,
>> the differences is mainly due to the new Partial Volume Estimation (PVE).
>> With this PVE model two additional mixed classes are estimated (GM-WM and
>> GM-CSF). The final results is estimated for the 3 main tissue classes
>> (GM/WM/CSF) and is now expressed as fraction (or amount) of each pure tissue
>> that is present in this voxel. The effect is mostly visible in the basal
>> ganglia, which are a mix of GM/WM. The final GM segmentation is around 50%
>> because these voxels also contain 50% WM. Although this value is lower
>> compared to non PVE approaches this is a more accurate segmentation.
>> For your data there are probably more voxels containing mixed classes.
>> Additionally, older segmentations with SPM5/VBM5 showed a tendency to
>> binarize the segmentations. Furthermore, the old approaches were based on a
>> bayesian approach with tissue priors of control subjects that often not
>> reflect the analyzed sample. The new segmentation algorithm in VBM8 does not
>> use any priors, thus for subjects that deviate from healthy control subjects
>> large differences in the segmentations might occur. However, your values for
>> GM/WM volume are quite different and I would rather check the raw volumes
>> that are saved in a text file.
>> Christian Gaser, Ph.D.
>> Assistant Professor of Computational Neuroscience
>> Department of Psychiatry
>> Friedrich-Schiller-University of Jena
>> Jahnstrasse 3, D-07743 Jena, Germany
>> Tel: ++49-3641-934752 Fax: ++49-3641-934755
>> e-mail: [log in to unmask]
>> On Fri, 28 May 2010 01:52:43 +0800, lion gao <[log in to unmask]> wrote:
>> >Dear Christian Gaser,
>> >Thanks for your great work on VBM8. It make segmentation easier.
>> >VBM5 or SPM5, SPM8 may fail to segment some T1 images (about 10%, e.g.
>> >strange C2. images), Now VBM8 seems work well and never fail to segment so
>> >One problem I find is that when I compare the images segmented by VBM5 and
>> >VBM8, they looks quite different. and indeed their volumes differ a lot.
>> >Mainly, the grey matter volume in VBM8 is smaller than VBM5, whereas the
>> >white matter volume is bigger. the TIV changes little. I segmented
>> >subjects from different groups. e.g. from young, old group and patient
>> >group, this trend is similar among the groups. please note that I use
>> >default setting in both VBM5 and VBM8, which are normalized modulated
>> >Attached pls find a figure for your reference. I wonder is there any
>> >for this. and which result may be better. BTW, another thing is that I
>> >the density in white matter from VBM8 can be 2.0778, which those in VBM5
>> >all less than 1.0. So what's the meaning of 2.0778.
>> >Would you please help me to understand these issues. Sorry that if you
>> >demonstrated these points in VBM8 manual. I have not read it thoroughly.
>> >Best wishes,