Katell,
Hello.
That is a very good point. What led me to the maximal classification is that I counted voxels in a tissue classification image instead of what you are saying and sum the actual modulated voxel values. I essentially created binary masks of each tissue type and then summed the voxels in the mask and multiplied by the voxel volume.
I like your argument regarding "border tissue" and if your end goal is TIV than what you say makes perfect sense. But if you want mask images ... then what do you do with multiple tissues occupying the same voxel with relatively high probabilities?
On another note, have you figured out a straightforward way to separate the ventricular CSF from the non-ventricular
CSF?
And are you using SPGRs for this?
Thank you,
Jason.
----- Original Message ----
From: mevel <
[log in to unmask]>
To:
[log in to unmask]Sent: Tuesday, June 5, 2007 10:03:00 AM
Subject: [SPM] TR: [SPM] Total intracranial volume: for information
Dear Jason,
thanks for your
comment.
This is an interesting issue that we did not
consider in our preliminary tests.
However, the question is then: Is it
really false to include a
voxel both in the GM and WM volumes
since its contribution is weighted by
its belonging percentage ? A
voxel could be at the boundary between WM and GM, and thus be associated with a
50% chance to be WM and 50% chance to be GM. As a result, it will be included in GM (with 0.5 ml
since we are in modulated data) and WM (another 0.5ml) volumes, but this would
make sense anyway. Indeed, it will
not be counted twice (e.g. adding
a 1 ml value in the two compartments for the same voxel) but as
two halves. Classifying a "bordering" voxel in only one of the two WM
and GM compartments seems to be also pretty false, doesn'
it?
All the
best,
Katell
De : Jason
Steffener [mailto:]
Envoyé : mardi 5 juin 2007
14:21
À : mevel; Objet : Re: [SPM] Total intracranial
volume: for information
Katell,
Bonjour.
I have looked into and performed similar analyses to make similar conclusions.
But what I also found was that by segmenting the brains, thresholding, combining
and then summing to calculate TIV some voxels are included twice. The
segmentation process gives probabilities so one voxel can with be 50% gray and
50% white and therefore be included twice with a threshold <0.5. What I did
find was that if the voxels are restricted to only being classified as a single
tissue type the thresholding makes much less difference. What I have done is to
classify voxels based on their maximal probability. Did you do this?
All
the best,
Jason.
-----
Original Message ----
From: mevel <
[log in to unmask]>
To:
[log in to unmask]Sent: Tuesday, June 5, 2007 5:43:13 AM
Subject: [SPM]
Total intracranial volume: for information
Hi
!
As expected in our previous post we have performed some
tests to assess the effect of different threshold methods and values on total
intracranial volume estimation from segmented and modulated MRI data
sets.
We
have used either different thresholds (from 0 = no threshold, to
0.9) applied on each of the brain partition, or differently thresholded
wholebrain masks (again using thresholds from 0 to 0.9). We thus calculated the
total intracranial volume of 19
healthy controls and 28
slightly atrophied patients in these 9*9 conditions, and assessed their
inter-correlations. As expected, all values were significantly correlated.
Highest correlations (r>0.97) were observed for thresholds from 0 to 0.5
(where TIV values decrease from 0 to 35%, respectively). Thresholds above this
value lead to decreased correlations (0.95 < r < 0.70) and more than
49%
decrease of the maximum TIV value.
We
thus conclude from this preliminary analysis that the use of <0.5 threshold
values would lead to more consistent TIV estimation, and this threshold could be
indifferently applied to either each brain partition or the whole brain
image.
Best
regards,
Katell
******************************************
Katell MEVEL
Doctorante
email :
[log in to unmask]
Tél : +33 (0)2 31 47 02
32
Neuropsychologie Cognitive
et Neuroanatomie Fonctionnelle de la Mémoire
Humaine
Equipe INSERM EMI0218 -
EPHE- Université de Caen Basse
normandie
Laboratoire de
Cycéron
Blvd Becquerel BP
5229
14074 Caen cedex
5
Tél : +33 (0)2 31 47 02
32
Fax : +33 (0)2 31 47 02
75
************************************************