Organize a benchmark dataset for fmri analysis

[gkiar]

Ideally including:

• Test retest data
• "High" quality samples
• "Typical" standards
• Heterogeneity

[gkiar]

Related to #3

[poldrack]

the NARPS dataset (https://openneuro.org/datasets/ds001734/versions/1.0.5) might be good for this...

…

On Mon, May 17, 2021 at 8:33 AM Greg Kiar ***@***.***> wrote: Related to #3 <#3> — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#4 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGUVEBA2NBBUGQTXIDP2RDTOEZMPANCNFSM44WTUCTQ> .

-- Russell A. Poldrack Albert Ray Lang Professor of Psychology Associate Director, Stanford Data Science Director, SDS Center for Open and Reproducible Science Building 420 Stanford University Stanford, CA 94305 ***@***.*** http://www.poldracklab.org/

[gkiar]

Thanks, @poldrack!

@TingsterX, is there a publicly available primate fmri dataset which could be used for benchmarking tools?

When thinking about #3, which focuses specifically on benchmarks for skull extraction on structural data, does anybody have an idea of an existing collection which has ground truth segmentations for either human or primate? I may push this question to the Twitter-verse...

[gkiar]

Also, cc: @hough129

[audreymhoughton]

I have started making a test dataset with the studies listed below with a few subjects each for something else (or maybe related to this) already. These datasets have / are going to have BIDS input, processed outputs from the abcd-hcp-pipeline, and derivatives (we haven't decided what those derivatives specifically entail).

These currently live on Box.

ABCD (one for each scanner type - two scanner types are ready)
HBN (5 subjects - almost ready - still uploading processed)
PNC (not processed yet - need to modify pipeline)
HCP-D (ready to go - two subjects)
NKI-Rockland (have not processed yet - need BIDS inputs)

[gkiar]

@hough129 for NKI data: http://fcon_1000.projects.nitrc.org/indi/enhanced/access.html

[arueter1]

Temporary storage location: S3 bucket.
Goals for future storage location: Loris study at MSI at UMN so that the raw dicoms / derivatives / subject data is easier to download whenever we want to share it. This won't be ready to go until later this year (Oct 2021?).

Action Items:
Are the datasets that Audrey listed above public? Any use considerations?
QSIPrep still needs to be run on some diffusion data.

[TingsterX]

For NHPs, check out the PRIME-DE (https://fcon_1000.projects.nitrc.org/indi/indiPRIME.html). The Oxford dataset has 20 macaques (~50min per monkey); UCdavis has 19 macaques, shorter fMRI scans but with higher-resolution of T1 and T2.

For the brain extraction dataset. Cameron had one human manually edited dataset open.
https://academic.oup.com/gigascience/article/5/1/s13742-016-0150-5/2737425?login=true

Recently, we just published a tool using a transfer-learning framework that trained the U-Net model on the human dataset and upgraded it with the macaque data. It also works for other species, e.g. chimps, marmosets, and pigs as well.
https://github.com/HumanBrainED/NHP-BrainExtraction

[gkiar]

Also adding @engfranco to the thread

[engfranco]

I have started making a test dataset with the studies listed below with a few subjects each for something else (or maybe related to this) already. These datasets have / are going to have BIDS input, processed outputs from the abcd-hcp-pipeline, and derivatives (we haven't decided what those derivatives specifically entail).

These currently live on Box.

ABCD (one for each scanner type - two scanner types are ready)
HBN (5 subjects - almost ready - still uploading processed)
PNC (not processed yet - need to modify pipeline)
HCP-D (ready to go - two subjects)
NKI-Rockland (have not processed yet - need BIDS inputs)

Folks, let me know if you have any questions about the NKI-Rockland or HBN datasets. If you need 5 good data from the NKI-Rockland dataset, I recommend using these 5 that have low motion:
sub-A00056703/ses-BAS1
sub-A00055906/ses-BAS1
sub-A00075732/ses-BAS1
sub-A00034073/ses-BAS1
sub-A00063006/ses-BAS1

Links to the S3 bucket of the whole imaging dataset organized in BIDS can be seen here:
http://fcon_1000.projects.nitrc.org/indi/enhanced/aws_links.csv

[arueter1]

Thanks Alexandre. One quick thing: @engfranco I'm not sure that we should have subject IDs out on a public facing website. Maybe we can share that internally to this team somehow (maybe via email?).

[gkiar]

For skull stripping, this looks awesome: http://preprocessed-connectomes-project.org/NFB_skullstripped/index.html

[gkiar]

Also from @TingsterX :

chimps: https://www.chimpanzeebrain.org
marmoset: https://brainatlas.brain.riken.jp/marmoset_html