I read with interest the comment by Lucas and some excellent responses by Anderson, and possibly others I did not read yet.

A. These are my personal views based on experience with T1-weighted, DTI-based, T2-weighted, multi-modal, combined T1-T2-PD-DTI tissue segmentation and lesion methods as we and others published on even across the human lifespan (see list of references below) using FreSurfer and other approaches. Voxel-based approaches could not be compared directly with volume-based results as you obtain from FS or FSL.

B. As summarized on the FreSurfer Wiki ( http://surfer.nmr.mgh.harvard.edu/fswiki), FreeSurfer has been validated extensively using postmortem data and one can list scores of human pathologies and hundreds of papers that utlilized FreeSurfer results including child brain development, aging, brain trauma, Schizophrenia, Bipolar disorders, Parkinson, Alzheimer's, multiple sclerosis, Autsim, dyslexia and even the effect of mother's milk on deevelopment. Recently one can also appreciate the use of FreeSurfer in perfusion mapping, Iron mapping (Hasan et al., MRM 2011) and the fusion of quantittaive MRI methods to study the interplay between metrics such as relaxation time and radial diffusivity (Walimini et al., 2011).

C. The problem is not in FreeSurfer results as you can always edit what you get or modulate it with experienced-rater masks or known landmarks. For example, thalamus in Fresurfer is a proper, and it is impossible to segment this region on T1-weighted alone into GM and WM. One can use T1-T2 and PD contrasts and hence you will get almost all gray matter zones in the thalmus along with its white matter islands. One can also use DTI along with cortical parcellations to regularize the solutions. For hippocampus, this is complicated and its subfields are highly variable, one may need higher spatial resolution methods, double fluid attenutaion and dedicated atalses. The Diencepahlon or brainstem are complicated as deep gray matter such as substantia nigra and red nucleus are not contrasted on T1-weighted data, so goes the Cerebellum and other regions (e.g. hypothalmus)

D. The problem shows up when one tries to compare results, but caution as volume results and methods (x, y) may be related by a scaling or shift (y=a*x+c). No segmentation or parcellation method that uses single modality will be able to segment with 100% accuracy CSF, GM and WM as the very defintion of boundaries is limited by the acquisition paradigm details. For example, total CSF can never be estimated on T1-weighted. Lesions look dark on T1-weighted, so does CSF and GM. If you have iron accumulation you may see it as bright on T1-weighted data which may be confounded as white matter, but the training set and raters may help dtect and remove this from your data.

E. From my experience, I am not a big fan of comparing results obtained from packages. I am more concerned with reliability, reproducibility on serial studies of one single package and sensitivity to variables such as Age, Side, Gender, Pathology and Clinical or Cognitive outcome ASSUMING your Scanner is STABLE and the groups being compared are carefully selected and screened (age range matched, comorbids, covariates)

F. Clearly in pathologies, plaques, tumors or lesions will confound any method used as hypo or hyper-intesnities will be interpreted differenctly and one has to intervene to overide the results or use other contrasts such as T2-wighted, or FLAIR.

G. As far as testing segmentation methods we made some suggestions in a recent Letter (Hasan and Pedraza Neuroimage 2009). I also bring attention for a paper that we published in August 2011 Neuroimage that used FreeSurfer and fused its regional GM and WM results with DTI-derived and Relaxation maps in both GM and WM. We did not find major problems with FreeSurfer volumetry if you carefully inspect the results, we have also tested subsets of the control data as function of age and compared it with known or expected trends using large samples.

H. Occasionallly we encountered cases with poor results, but be careful. Failure or poor results may not be fairly blamed on FreeSurfer as some cases had poor SNR, some motion artifcats and we removed them from the analysis. FreeSurfer results are as good as your input data contrast-to-noise, slice thickness and other image acquisition factors that control the contrast (e.g. flip, TE/TR,..). I wont expect magic if the data original contrast and fed to any program does not have sufficient signal and contrast between CSF, GM and WM.

I. References

1. http://surfer.nmr.mgh.harvard.edu/fswiki

2. Hasan KM, Frye RE. Diffusion tensor-based regional gray matter tissue segmentation using the international consortium for brain mapping atlases. Hum Brain Mapp. 2011;32(1):107-17.

3. Hasan KM, Walimuni IS, Kramer LA, Frye RE. Human brain atlas-based volumetry and relaxometry: application to healthy development and natural aging. Magn Reson Med. 2010;64(5):1382-9.

4. Hasan KM, Walimuni IS, Kramer LA, Narayana PA. Human brain iron mapping using atlas-based T(2) relaxometry. Magn Reson Med. 2011 (doi: 10.1002/mrm.23054).

5. Hasan KM, Walimuni IS, Abid H, Hahn KR. A review of diffusion tensor magnetic resonance imaging computational methods and software tools. Comput Biol Med. 2011 (doi:10.1016/j.compbiomed.2010.10.008)

6. Hasan KM and Pedraza O. Improving the reliability of manual and automated methods for hippocampal and amygdala volume measurements. Neuroimage. 2009;48(3):497-8.

7. Walimuni IS, Hasan KM. Atlas-based investigation of human brain tissue microstructural spatial heterogeneity and interplay between transverse relaxation time and radial diffusivity. Neuroimage. 2011;57(4):1402-10.

8. Walimuni IS, Abid H, Hasan KM. A computational framework to quantify tissue microstructural integrity using conventional MRI macrostructural volumetry. Comput Biol Med. 2011 (doi:10.1016/j.compbiomed.2010.10.009).

Khader M Hasan, PhD Associate Professor of Radiology MSE 168, Tel 713 500 7690 (FAX 713 500 7684) University of Texas Health Science Center at Houston Medical School Diagnostic and Interventional Imaging Magnetic Resonance Imaging Research Division Diffusion Tensor Imaging Lab, Tel 713 500 7683 http://www.uth.tmc.edu/radiology/faculty/khader-m-hasan/index.html ________________________________________ From: freesurfer-bounces@nmr.mgh.harvard.edu [freesurfer-bounces@nmr.mgh.harvard.edu] On Behalf Of freesurfer-request@nmr.mgh.harvard.edu [freesurfer-request@nmr.mgh.harvard.edu] Sent: Monday, August 29, 2011 11:00 AM To: freesurfer@nmr.mgh.harvard.edu Subject: Freesurfer Digest, Vol 90, Issue 93

Send Freesurfer mailing list submissions to freesurfer@nmr.mgh.harvard.edu

To subscribe or unsubscribe via the World Wide Web, visit https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer or, via email, send a message with subject or body 'help' to freesurfer-request@nmr.mgh.harvard.edu

You can reach the person managing the list at freesurfer-owner@nmr.mgh.harvard.edu

When replying, please edit your Subject line so it is more specific than "Re: Contents of Freesurfer digest..."

Today's Topics:

1. Re: Accuracy of FreeSurfers gray matter segmentation (Anderson Winkler) 2. Re: Accuracy of FreeSurfers gray matter segmentation (Lucas Eggert) 3. patch present-label missing (maahmed@mappi.helsinki.fi) 4. Re: Paired analysis results into table? (Braber, A. den) 5. Re: patch present-label missing (Bruce Fischl) 6. Re: Accuracy of FreeSurfers gray matter segmentation (Anderson Winkler) 7. Re: Accuracy of FreeSurfers gray matter segmentation (Bruce Fischl) 8. FW: Cortical Flattening (Tax, C.M.W.) 9. Re: problem with selxavg3 (Douglas N Greve) 10. Re: problem with selxavg3 (Maryam Vaziri Pashkam) 11. Re: problem with selxavg3 (Douglas N Greve) 12. Re: Correlation cortical thickness and lgi (Douglas N Greve)

----------------------------------------------------------------------

Message: 1 Date: Sun, 28 Aug 2011 16:47:18 -0400 From: Anderson Winkler andersonwinkler@hotmail.com Subject: Re: [Freesurfer] Accuracy of FreeSurfers gray matter segmentation To: freesurfer@nmr.mgh.harvard.edu Message-ID: BLU0-SMTP351AC314B26178D2AADB33DA1150@phx.gbl Content-Type: text/plain; charset="iso-8859-1"

Hi Lucas,

To know if a given structure is gray or white matter you can look in any reasonable anatomy textbook. In any case, the question itself is somewhat ill-posed, because some of the subcortical structures have heterogeneous tissue composition and can't really be labeled entirely as gray matter, even macroscopically. The most notable examples are perhaps the thalamus and hippocampus, but the same applies to other structures too.

Anyway, if you really want to make a hard distinction, you can call then caudate, putamen, pallidum, amygdala, accumbens, hippocampus and thalamus as gray matter. The region defined as ventral diencephalon is very heterogeneous and I would not classify it either as GM or WM, as it includes mamillary bodies, tuber cinereum/infundibulum (but not hypophysis), some hypothalamic nuclei near the lateral and inferior walls of the 3rd ventricle and sometimes fragments of the optic tracts (but not chiasm, which has its own label). It also includes parts of the mesencephalon (e.g. part of the cerebral crux, part of the substantia nigra and rubra).

Importantly, if you are comparing algorithms, you have to be sure they are reporting the same thing. For instance, it's fairly common to run SPM or FSL/FAST segmentation, then sum the GM voxels within a region defined from an atlas. If you do this for, say, caudate or thalamus, you'll get the volume of what the algorithm classified as GM within the structure you selected. FreeSurfer (and, e.g. FSL/FIRST), on the other hand, will segment and report the volume of the structure as a whole, including all what it contains. A direct comparison, thus, is not valid.

Hope this helps!

All the best,

Anderson

On 28/08/11 02:53, Lucas Eggert wrote:

Dear List,

I recently posted the message you see below.

I currently compare the segmentation results of different segementation algorithms and FreeSurfer always yields the least accurate results. Because I still have difficulties to decide which of the labels in the aseg.mgz files should be considered gray matter, I have the feeling, the bad results I am getting for FreeSufer might be caused by not including all relevant gray matter segments.

So, I would very much appreaciate any help on deciding which of the labels in the aseg.mgz file belong to gray matter!

With kind regards Lucas Eggert

-------- Original-Nachricht -------- Betreff: [Freesurfer] How to determine gray matter in the aseg.mgz Datum: Tue, 03 May 2011 17:26:55 +0200 Von: Lucas Eggert leggert@Uni-Osnabrueck.DE An: freesurfer@nmr.mgh.harvard.edu freesurfer@nmr.mgh.harvard.edu

Dear Experts,

I would like to generate a gray matter mask using the aparc+aseg.mgz file.

Now, my question is whether there exists a file in addition to the FreeSurferColorLUT.txt file, which more explicitly, i. e., withouth the abbreviations, explains the labels, because I find it hard to decide whether, e. g., "Left-VentralDC", "Line1", "LeftmOg", or "Left-Interior" belong to gray matter, or not (to name only a few of those labels I cannot clearly classify).

Or is there a easier way to create a gray matter mask?

Thank you very much in advance!

Best regards, Lucas Eggert

-- Lucas Eggert, M.Sc. Institute of Cognitive Science University of Osnabrueck Albrechtstrasse 28 D-49076 Osnabrueck Germany

Phone: +49-541-969-44-28 Website:http://www.cogsci.uni-osnabrueck.de/~leggert/

---

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer

The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.