Hi Meng,

here are some reasons for different looking images (and therefore different structural estimates): A Acquisition A1. Head Motion during acquisition: See e.g. http://reuter.mit.edu/publications/pid/reuter-motion14 http://reuter.mit.edu/publications/pid/tisdall15

A2. Hydration levels http://reuter.mit.edu/publications/pid/biller15

A3. different head positions can cause large differences if images are not gradient unwarped (depending on your hardware these effects can be small or large). Also there is of course noise in the images which can lead to different results.

B Then there are reasons in the processing

B1. Different noise or small changes in the image can trigger processing to do different things (e.g. small skull strip failure, different surface placement, eg. one time including dura the other time not). This can produce arbitrary large changes in a single subject. It can be reduced/fixed via manual edits

B2. Processing bias, for example, mapping the follow up to baseline (which is what I think you do below), is problematic as follow ups get interpolated (which looks like smoothing) and so you introduce a change to some of your images, biasing your measurements of change. see http://reuter.mit.edu/publications/pid/reuter-long12 http://reuter.mit.edu/publications/pid/reuter-bias11 This problem (and some of B.1) can be improved by using the longitudinal pipeline in FreeSurfer to analyze your data: https://surfer.nmr.mgh.harvard.edu/fswiki/LongitudinalProcessing

Also from your text below it seems like you are not processing the original images, but you first map them to fsaverage ('further normalized into fsaverage space')? You should run FreeSurfer on the original input images (using the longitudinal stream).

Finally note, that you cannot trust individual cortical thickness values. People usually run studies comparing 10 or 15 subjects per group. There can always be individual outliers. Manual checking and editing can fix some of that, but not all. Especially effects from motion or hydration cannot be removed.

Best, Martin

On 04/07/2016 12:02 PM, Meng Li wrote:

​Hi everyone,

I am trying to extract regional cortical thickness but confused with the altered values for the same subject within two days.

In our experiment, participants were scanned with the identical protocol in three times continuously: including baseline, 2 hours, 24 hours later. All T1s were processed by FS 5.3 and further normalized into fsaverage space, as well as a customized ROI. Interestingly, both of the regional CT and average CT of whole brain showed impressive variance among three timepoints for the same subjects. Moreover, if the T1 in the second and the third scan was coregister to the baseline scan and rerun the pipeline, the values is not very convincing. I wonder this variance was caused by the different head position in the scanner. Is there any suggestion to get rid of that? Which cortical thickness value could be trust most in my case? Any comments is appreciated. Thank you in advance.

​Here is one example from my test. ​Subject regional CT average CT of whole brain bu45_t1 2.49833562 tel:49833562 2.18242687 bu45_t2 2.540465325 2.233703478 bu45_t3 2.63002383 2.202483542 bu45_t2_to_t1 2.636101336 2.237099006 bu45_t3_to_t1 2.499173829 tel:499173829 2.216584361

All the best Meng

---

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