Data sharing in neuroimaging: incentives, tools, and challenges

Chris Gorgolewski Department of Psychology Stanford University

HOW CAN YOU BENEFIT FROM DATA SHARING?

NKI Enhanced

•  329 subjects (will reach 1000) – Representative sample: young and old, some with

mental health history •  1 hour worth of MRI (3T) scanning: – MPRAGE (TR = 1900; voxel size = 1mm isotropic) –  3x resting state scans (645msec, 1400msec, and

2500msec) – Diffusion Tensor Imaging (137 direction; voxel size

= 2mm isotropic) – Visual Checkboard and Breath Holding

manipulations  

fcon_1000.projects.nitrc.org/indi/enhanced/

Human Connectome Project •  > 500 subjects (will reach 1200)

–  Young and healthy (22-35yrs) –  200 twins!

•  1 hour worth of MRI scanning: –  State of the art sequences – high temporal and spatial resolution –  Resting-state fMRI (R-fMRI) –  Task-evoked fMRI (T-fMRI)

•  Working Memory •  Gambling •  Motor •  Language •  Social Cognition •  Relational Processing •  Emotion Processing

–  Diffusion MRI (dMRI) –  MEG and EEG –  7T coming soon  

Human Connectome Project

•  Rich phenotypical data – Cognition, personality, substance abuse etc.

•  Genotyping! (not yet available)

•  Methodological developments – Fine tuned sequences –  Innovative field inhomogeneity corrections – New preprocessing techniques

•  Ready to use preprocessed data

humanconnectome.org

FCP/INDI Usage Survey

Survey Courtesy of Stan Colcombe & Cameron Craddock

FCP/INDI Data Usage Description        Master's thesis research 11.94% Doctoral dissertation research 38.81% Teaching resource (projects or examples) 13.43% Pilot data for grant applications 16.42% Research intended for publication 76.12% Independent study (e.g., teach self about analysis) 37.31%

FCP/INDI Users; 10% respondent rate

Growth of the reuse of OpenfMRI datasets

Motivation

•  Share  your  stat  maps!  

vs.

institutions scientists

Data sharing saves money

$878,988 cost of reacquiring data for each of the

reuses of OpenfMRI datasets

Data sharing fears

•  Fear of being scooped •  Fear of someone finding a mistake •  Misconceptions about the ownership of the

data

Studies sharing data have higher statistical quality

Wicherts JM, Bakker M, Molenaar D (2011) Willingness to Share Research Data Is Related to the Strength of the Evidence and the Quality of Reporting of Statistical Results. PLoS ONE 6(11): e26828. doi: 10.1371/journal.pone.0026828

Neuroimaging data sharing hierarchy

Poldrack and Gorgolewski, 2014

Just coordinates?

•  Databases such as Neurosynth or BrainMap rely on peak coordinates reported in papers (only strong effects)

Are we throwing money away?

Baby steps

•  Everything is a question of cost and benefit –  If we keep the cost low even small benefit (or

just conviction that data sharing is GOOD) will suffice

NeuroVault.org simple data sharing

•  Minimize the cost! •  We just want your statistical maps with

minimum description (DOI) –  If you want you can put more metadata, but

you don’t have to

•  We streamline login process (Google, Facebook)

NeuroVault.org

Gorgolewski, et al., submitted

Benefits - visualisation

Benefits - decoding

Live demo

Benefits - other

•  Private collections •  Multiple contributors to one collection •  Sharable persistent URLs •  Viewer embeddable on your labs website

or your private blog •  Improved exposure of your research •  Improved reusability of your results

Using NeuroVault…

•  Improves collaboration •  Makes your paper more attractive •  Shows you care about transparency •  Takes only five minutes •  Gives you warm and fuzzy feeling that you

helped future meta-analyses

Validation and gains in sensitivity

NeuroVault for developers

•  RESTful API (field tested by Neurosynth) •  Source code available on GitHub

What is NIDM-Results?

Neuroimaging data sharing hierarchy

Poldrack and Gorgolewski, 2014

MAKING DATASHARING COUNT Credit where credit’s due

Quality control

•  Share  your  stat  maps!  

Complex datasets require elaborate descriptions

•  Share  your  stat  maps!  How can we appropriately reward extra effort and risk related with sharing data?

Solution – data papers

•  Authors get recognizable credit for their work. – Even smaller contributors such as RAs can be

included.

•  Acquisition methods are described in detail.

•  Quality of metadata is being controlled by peer review.

Gorgolewski, Milham, and Margulies, 2013

•  Neuroinformatics (Springer) •  GigaScience (BGI, BioMed Central) •  Scientific Data (Nature Publising Group) •  F1000Research (Faculty of 1000) •  Data in Brief (Elsevier) •  Journal of Open Psychology Data (Ubiquity

press)

Where to publish data papers?

What makes a good data paper?

•  Clear and accurate description of the acquisition protocol.

•  Good data organization. •  Ease of access to data. •  Data quality description. •  Fair credit attribution.

How to improve the impact of your dataset?

•  Provide preprocessed data. •  Reach out to your peers… – …and people outside of your field (ML)

•  Build a community around the data.

StudyForrest.org

Repositories

•  Field specific – OpenfMRI.org (task based fMRI) – FCP/INDI (resting state fMRI) – COINS

•  Field agnostic – DataVerse (Harvard) – Figshare (only small datasets) – DataDryad (fees may apply)

OpenfMRI

•  Will host any dataset that has a task based fMRI component

•  No fees •  Curated and uncurated datasets •  Recommended by many journals (including

Scientific Data)

Prepare in advance

•  Make sure your consent form includes data sharing

•  Decide which database you want to send your data to in advance – Organize your data according to their

requirements

•  Work on anonymized data as much as you can

If I haven’t convinced you yet

•  Why to share data: –  It’s the ethical thing to do (Brakewood and

Poldrack 2013) – The journal might require it (PLoS). – Your funders might require it (NIH). – Track record of data sharing can improve your

chances of getting your next grant.

Sharing data is related to higher citation rate

Piwowar, Day & Fridsma (2007)

Piwowar & Vision(2013)

Acknowledgements

Russell A. Poldrack Jean-Baptiste Poline

Yannick Schwarz Tal Yarkoni

Michael Milham Daniel Margulies

Yannick Schwartz Gael Varoquox

Joseph Wexler Gabriel Rivera Camile Maumet Vanessa Sochat Thomas Nichols MPI CBS Resting state group Poldrack Lab INCF Data Sharing Task Force