Thanks for your good points, Antonin.

More on the matter:

When segmenting standard resolution (~ 1mm) T1 scans, the position of the internal boundaries between the hippocampal substructures largely relies on prior knowledge acquired from our ex vivo training data and summarized in our statistical atlas. Therefore, the volumes of subfields around these boundaries (especially internal subfields such as the GC-DG and molecular layer) must be interpreted with caution, and further validation with higher resolution data should be carried out to confirm the results.

 Having said that:

1. The fimbria is visible in 1 mm T1 data. It is true that, being a thin structure, its volume might be affected by motion artifacts more than that of other structures (see our Neuroimage paper linked by Antonin).

2. The hippocampus / amygdala interface is given by the alveus, which is a very thin structure, yielding a very faint (but not invisible) boundary in 1mm data, particularly if the scan is affected by motion. For that reason, automated methods sometimes make mistakes in that area, but they mostly get it right, partly thanks to the prior knowledge encoded in atlases.

3. The tail relies on geometric criteria for the anterior / posterior cut-off, but does not attempt to infer any internal boundaries, so it is definitely reliable even when segmented from 1 mm scans.

Jerome, for your specific case:

- The tail should not be problematic, for reason 3 above.
- Schoemaker’s article, in addition to using an ancient version of FreeSurfer,  is about segmentation of infant MRI, which is more difficult that segmenting adults.
- Volumetric protocol: see Table 2 of our Neuroimage paper describing the hippocampal atlas. For the tail: “we identified the first coronal slice (anterior to posterior) where the fornix is fully connected to the hippocampus, and labeled the whole hippocampus with this “umbrella” label in the remaining slices”.
- The fimbria is visible (reason 1 above).
- The molecular layer is not visible at 1 mm resolution, but you can still get and idea of its volume thanks to the prior - again, results based on such volumes should be interpreted with caution. However, if I understood correctly, your results are on the tail, and hence reliable (point 3).

I hope this helps.

Kind regards, and sorry for the long response!

/Eugenio

Juan Eugenio Iglesias

ERC Senior Research Fellow
Translational Imaging Group
University College London
http://www.jeiglesias.com
http://cmictig.cs.ucl.ac.uk/

On 13 Jan 2017, at 01:23, Antonin Skoch <ansk@ikem.cz> wrote:

Dear Jerome,

I think I could chime in with my opinion: The Schoemaker et al paper results are based on hippocampus/amygdala segmentations in FreeSurfer 4.4.

Hippocampal subfields segmentation module from development version you used should be more precise than 4.4. version and also the anatomical landmarks are well specified in paper Iglesias et al, 2015:

https://www.ncbi.nlm.nih.gov/pubmed/25936807
I think this paper can be used as a good basis for arguments to your reviewer.

Concerning the reliability of estimation of particular hippocampal subfields, I think it also depend on what resolution was your input images, and whether you used additional high-resolution T2. The effect of various image resolution on the reliability of results is also well discussed in above-mentioned paper I think.

Regards,

Antonin Skoch

_______________________________________________
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.