Difference between DKT40 and DKT31 OASIS?

[WNZhao1988]

Dear Mindboggle project maintainers,

Can I assume DKT40 in Freesurfer and DKT31 OASIS (OASIS-TRT-20_jointfusion_DKT31_CMA_labels_in_MNI152_v2.nii.gz) are the same, except one is the surface version and the other is the volumetric version? Since after I run reconall in Freesurfer and get the DKT40 outputs, I also need to do some analysis in volumetric space. So, I want to ensure the parcellation I use in volumetric MNI space is the same as the parcellation DKT40 in Freesurfer.

Best regards,
Naren

[sgiavasis]

Hi Naren @WNZhao1988,

Off-hand, I'm not entirely sure if they are exactly the same. DKT40 and DKT31 both have the different numbers of parcellations. From what I understand, it is true that DKT40 is better suited for surfaces, and DKT31 better suited for volumetric stuff (I believe DKT31 was created with a combination of landmarks but also probabilistic maps, while DKT40 was only landmarks).

But I'm not sure if or how DKT40's 40 parcellations map directly to DKT31's 31 parcellations.

When you say you need to do some analysis in volumetric space, is this independent of the processing you did in Freesurfer, or are you using some of the Freesurfer outputs to inform/convert over to volumetric space?

[WNZhao1988]

Hi Sgiavasis @sgiavasis,

Thank you for your reply! I think DKT40 and DKT31 may be the same because it seems parcels in DKT31 are all covered in DKT40 based on the parcel labels. The analysis in volumetric space is this independent of the processing in Freesurfer. Basically, I want to find the same brain region in volumetric space with DKT31 and in surface space with DKT40.

[binarybottle]

For the DKT31 labeling protocol, please see: https://mindboggle.readthedocs.io/en/latest/labels.html

The DKT40 is a brain atlas created from 40 brains labeled according to the DKT31 labeling protocol. Please see:
https://www.frontiersin.org/articles/10.3389/fnins.2012.00171/full

FreeSurfer’s DK classifier atlas assigned the initial labels for 54 of the brains in the Mindboggle-101 data set (OASIS-TRT-20, HLN-12, MMRR-21, and MMRR-3T7T-2). These were then manually edited by GM and JT to conform to the DKT protocol as described above. We selected the first 40 brains that we labeled (20 male, 20 female, 26 ± 7 years of age, from the MMRR-21, OASIS-TRT-20, and HLN-12 data) to train a new FreeSurfer cortical parcellation atlas representing the DKT protocol (see http://surfer.nmr.mgh.harvard.edu/fswiki/FsTutorial/GcaFormat; S’egonne et al., 2004; Desikan et al., 2006 for details regarding the algorithm that generates the atlas and how it is implemented). The resulting “DKT40 classifier atlas” then automatically generated the initial set of cortical labels for the remaining 47 brains in the data set (see http://surfer.nmr.mgh.harvard.edu/fswiki/mris_ca_label). To our knowledge, the DKT40 atlas was generated in the same manner as the DK atlas except for differences in the labeling protocol and training set.