Yea, I would not say it is the best way to do QC. But just correlating the volumes themselves should be meaningful
On 9/9/2020 10:29 AM, Chris Vriend wrote:
External Email - Use Caution
Dear Eugenio,
I tested it on 103 subjects. The asymmetry index is significantly different from 0 (symmetric) for both aseg and ThalamicNuclei.v12. Also when you compare the asymmetry index across segmentation methods (either using paired or independent t-tests) the AI is significantly different. The correlation of the AI between methods is also 'only' r = 0.377 ( P < 0.001). I've attached an xls file with the data. Based on the reply from Douglas and yourself, does this mean that AI for aseg stems from both the method and data, while the asymmetry shown by your method is due to the data only and therefore comparing the AI between the two methods is meaningless and definitely not suitable as a quality (or sanity) check?
Cheers, Chris
Op di 8 sep. 2020 om 09:05 schreef Chris Vriend <chrisvriend@gmail.com mailto:chrisvriend@gmail.com>:
Dear Eugenio, Douglas referred me to you for this question. Do you have an explanation for the difference in asymmetry between the native (aseg) and ThalamicNuclei.v12 segmentation? kind regards, Chris /I'm not sure about this. The FS segmentation atlas was not created to be symmetrical, so interpreting the asymmetry of aseg results can be tricky. I know that Eugenio often creates his atlases to be symmetric; unfortunately, he's away until next week. Thalamus is also quite tricky because the contrast with WM is so low it can make finding the border quite variable. You are right about LGN/MGN. So, try re-sending this next week and see what Eugenio has to say./ Op ma 31 aug. 2020 om 17:21 schreef Chris Vriend <chrisvriend@gmail.com <mailto:chrisvriend@gmail.com>>: Dear freesurfer experts, I’m currently using FreeSurfer 7.1 with the thalamus subsegmentation from Iglesias et al (version 12) to subsegment the thalamus. Something we noticed is that the asymmetry between the left and right whole thalamus is reversed when comparing the native thalamus segmentation by FreeSurfer itself and the Iglesias method. This is exemplified by the values below where we calculated the Asymmetry Index [ (Left – right)/(left + right) * 100 ] and shows that for some subjects the left thalamus is larger when looking at the results of the Iglesias method, but smaller with the native method. To allow comparison between the two methods we subtracted the LGN and MGN volumes from the whole thalamus volume, because – if I’m not mistaken – these nuclei are not segmented by the standard recon-all pipeline? We don’t know why we observe this and not just in one dataset or one subject but in multiple. Do you have any words of wisdom or explanation for this phenomenon? Your advice is much appreciated. Kind regards, Chris Vriend Whole_thalamus_lh_Iglesias whole_thalamus_rh_Iglesias AI Fsnative_Left-Thal Fsnative_Right-Thal AI 4913.391 5129.889 -2.155650345 5987.2 5764.6 1.894177913 6019.185 6235.017 -1.761289719 6793.9 6732.4 0.454669791 6374.575 6574.893 -1.546920692 7543.9 7592.1 -0.318446089 6974.051 6983.314 -0.066366395 7661.3 7624.8 0.238779021 5907.853 5798.037 0.93812602 6756.4 6485.2 2.048090865 6316.792 6382.639 -0.518503546 7191.2 6959.7 1.63593835
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer