Bug fixed in Schaefer 2018 parcellation label names

[Valeria Kebets]

Dear CBIG repo users,

Thanks to feedback from Annie Gupta and her team, we found out that some of the Schaefer parcels have the wrong label names, although the parcel boundaries are correct. For example, in the 400-region parcellation, all parcels corresponding to Somatomotor Network B are labelled as "auditory" (i.e., 17Networks_LH_SomMotB_Aud_x), but some of them are obviously not in auditory cortex.

This error arose because of a bug in the algorithm that matches the Schaefer parcellations with the Yeo2011 split components. Note that the Yeo2011 split components came from splitting the original Yeo2011 7/17 networks into spatially contiguous components (https://github.com/ThomasYeoLab/CBIG/tree/master/stable_projects/brain_parcellation/Yeo2011_fcMRI_clustering). On closer inspection, we also felt that the names of a small number of the Yeo2011 split components were not appropriate.

We have resolved the issues and the updated label names have been released (Release version 0.14.3: https://github.com/ThomasYeoLab/CBIG/releases/tag/v0.14.3-Update_Yeo2011_Schaefer2018_labelaname). See below for more details about label names in Schaefer parcellations.

Parcels' label names

Parcel names are obtained by concatenating the number of networks, hemisphere, network name, component name, and parcel number.
e.g., for 17Networks_LH_DefaultA_pCunPCC_1:

• 17Networks means that this parcellation has been matched to the Yeo2011 17 networks
• LH means that this parcel is in the left (L) hemisphere (H)
• DefaultA is the name of the network that the parcel belongs to
• pCunPCC is the component name

Component names are sometimes coarse because they were generated from the Yeo2011 split component parcellation (7/17 networks), where parcels are of relatively low resolution (51/114 regions).

Generating parcels' label names in Schaefer parcellations

We developed a new matching algorithm and re-labeled some of the Schaefer parcels. The parcels whose label names have been changed are documented in the following .csv files that provide more details:

• ${CBIG_CODE_DIR}/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal/Parcellations/Updates/Update_20190916_component_changes.csv
• ${CBIG_CODE_DIR}/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal/Parcellations/Updates/Update_20190916_network_changes.csv

Our updated algorithm does the matching in two steps:

1. The Schaefer parcels are first matched to one of Yeo2011 7/17 networks. Each parcel from Schaefer parcellations is then assigned to a network, e.g., 17Networks_LH_SomMotB.
2. After that, each Schaefer's parcel is matched to one of the Yeo2011 split components and is assigned a component name, e.g., 17Networks_LH_SomMotB_Cent.

IMPORTANT NOTES

1. When we match the high-resolution Schaefer parcellation (e.g., 1000 parcels) to the Yeo2011 split components, the name might not fully describe each parcel. For example, a Yeo2011 split component might have the name "ParOcc" because it overlaps across parietal (Par) and occipital (Occ) regions. A high-resolution Schaefer parcel might match to the split component and the resulting name is 17Networks_LH_DorsAttnA_ParOcc_x, but this Schaefer parcel might only be in occipital cortex.

2. Component names are more detailed in the 17 networks than in the 7 networks parcellations.

3. Our lab generally uses the 400-region Schaefer parcellation matched to the Yeo2011 17 networks.

Details about changes

Changes in parcel ordering
For most parcels that were incorrectly labeled, only the component name has changed. However, for a smaller number of parcels, the network name has also been changed. Note that NOT all resolutions are affected. For parcellations in which only the component names have changed (but the network names stayed the same), we have kept the same ordering of the parcels (in the annot file, colortable, etc.) for the sake of consistency. However, for parcellations in which the network names have also changed, then we re-ordered the parcels.

• Parcellations whose network names are the same but some component names might have changed

For example, for the 100-region Schaefer parcellation, 7Networks_LH_DorsAttn_FEF_1 was changed to 7Networks_LH_DorsAttn_PrCv_1. The table below lists the parcellations that were affected by changes:

Resolution	# Parcels affected
100 Parcels, 7 Networks	16
200 Parcels, 7 Networks	29
300 Parcels, 7 Networks	42
400 Parcels, 7 Networks	58
500 Parcels, 7 Networks	67
600 Parcels, 7 Networks	87
700 Parcels, 7 Networks	97
800 Parcels, 7 Networks	112
1000 Parcels, 7 Networks	123
100 Parcels, 17 Networks	18
200 Parcels, 17 Networks	36
300 Parcels, 17 Networks	50
400 Parcels, 17 Networks	70

• Parcellations whose network names have changed for some parcels

The network name was changed for a small number of parcels in higher-resolution Schaefer parcellations (usually those near the network boundaries). For example, for the 900-region Schaefer parcellation, 7Networks_LH_DorsAttn_FEF_1 has been changed to 7Networks_LH_Cont_PFCd_1. The table below lists the parcellations that were affected by changes:

Resolution	# Parcels whose component name changes	# Parcels whose network name changes
900 Parcels, 7 Networks	133	1
500 Parcels, 17 Networks	92	1
600 Parcels, 17 Networks	115	2
700 Parcels, 17 Networks	158	2
800 Parcels, 17 Networks	157	1
900 Parcels, 17 Networks	209	4
1000 Parcels, 17 Networks	225	2

NOTE ABOUT THE LIMBIC NETWORK
The Limbic A/B networks in the 17-Network parcellations were displayed in different colors, but the label names did not distinguish between A/B. These labels have now been changed to provide the Limbic A/B distinction.

Mapping parcellation ordering between versions

For those who were using the old Schaefer parcellations, we provide a .mat file that contains the mapping vector between the old and the updated Schaefer parcellations:

• ${CBIG_CODE_DIR}/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal/Parcellations/Updates/Update_20190916_index_changes.mat

This .mat file is a 20 x 3 cell matrix, where 20 corresponds to 20 parcellation resolutions. The first column is the parcellation resolution, the second column is the #parcel x 1 mapping vector, and the third column indicates whether the ordering of the parcellation has been changed.

If users want to generate the mapping vector on their own, we also provide a script:

• ${CBIG_CODE_DIR}/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal/Parcellations/Code/CBIG_gwMRF_index_trans_btwn2versions.m 

How to update your repo

We have pushed a release to the public repository.

1. If you previously cloned or forked our public repo (https://github.com/ThomasYeoLab/CBIG), you need to update the repo by running:

cd $CBIG_CODE_DIR
git checkout master
git pull origin master

2. If you previously cloned our private repo (https://github.com/YeoPrivateLab/CBIG_private), i.e., Level-1 user of our private repo, you need to do:

cd $CBIG_CODE_DIR
git checkout develop
git pull upstream develop

3. If you previously forked our private repo, i.e., Level-2 user of our private repo, you also need to do the steps in 2. In addition, after finishing the above commands, you also need to update your forked remote repo by running:

git push origin develop

We apologize for the inconvenience caused!

CBIG lab

[Valeria Kebets]

Apologies, the correct GitHub link to the update release is here:

https://github.com/ThomasYeoLab/CBIG/releases/tag/v0.14.3-Update_Yeo2011_Schaefer2018_labelname

--

Valeria Kebets, PhD

Computational Brain Imaging Group

National University of Singapore

[Jo Etzel]

FYI, I'm seeing 66 parcels with component name changes for 400Parcels_7Networks, rather than the 58 listed in the table above. The additional ones with changed names are 135:142, as in this screen grab (the right side is the history for CBIG/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal/Parcellations/HCP/fslr32k/cifti/Schaefer2018_400Parcels_7Networks_order_info.txt). I haven't compared any others.

[Thomas Yeo]

Hi Jo,

Thanks! We will double check and get back to you.

Thanks,

Thomas 

On Tue, Sep 17, 2019 at 2:48 AM Jo Etzel <j.a....@email.wustl.edu> wrote:

FYI, I'm seeing 66 parcels with component name changes for 400Parcels_7Networks, rather than the 58 listed in the table above. The additional ones with changed names are 135:142, as in this screen grab (the right side is the history for CBIG/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal/Parcellations/HCP/fslr32k/cifti/Schaefer2018_400Parcels_7Networks_order_info.txt). I haven't compared any others.

--
You received this message because you are subscribed to the Google Groups "CBIG users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cbig_users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cbig_users/7ed1ecf6-2523-449a-ab78-5ca7074eac45%40googlegroups.com.

[Thomas Yeo]

Hi Jo,

We figured out the reason. For some parcels, name might change from "XXX_Cont_PFCl_2" to "XXX_Cont_PFCl_1". Because "Cont_PFCl" stayed the same, we did not mention this parcel has having different network or component names. But you are right that some users might want to know the numbers have changed from 1 to 2. 

We will add another column in the readme and we will also provide another csv table for numbering changes. Will email the list when this is done.

Apologies for the inconvenience, and please keep the feedback flowing :)

Thanks,
Thomas

On Tue, Sep 17, 2019 at 2:48 AM Jo Etzel <j.a....@email.wustl.edu> wrote:

FYI, I'm seeing 66 parcels with component name changes for 400Parcels_7Networks, rather than the 58 listed in the table above. The additional ones with changed names are 135:142, as in this screen grab (the right side is the history for CBIG/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal/Parcellations/HCP/fslr32k/cifti/Schaefer2018_400Parcels_7Networks_order_info.txt). I haven't compared any others.

--