Upload
camille-maumet
View
420
Download
0
Embed Size (px)
DESCRIPTION
Due to the lack of data shared when reporting neuroimaging results, most neuroimaging meta-analyses are based on peak coordinate data. However, the best practice is an image-based meta-analysis that combines full image data of the effect estimates and standard errors derived from each study. The Neuroimaging Data Model (NIDM) is an ongoing effort, supported by the INCF, to provide a domain-specific extension of the W3C PROV-DM. In this talk, I will review our recent progress in extending NIDM to share the statistical results of a neuroimaging study and our interactions with existing software packages (SPM, FSL, AFNI, Neurovault.org).
Citation preview
University of Warwick, Neuroimaging Statistics, Coventry, UK.
Camille Maumet
Supporting image-based meta-analysis with NIDM: Standardized reporting of neuroimaging results
INCF NIDASH Task force
Oct. 8th 2014
Agenda
• Context – NIDM and the INCF NIDASH Task force – Meta-analysis use-case – Data sharing environment
• NIDM for meta-analysis – NIDM-Results – Implementation – Future directions
• Conclusions
2
CONTEXT
3
INCF NIDASH Task Force CONTEXT
4
5
International Neuroinformatics Coordinating Facility
Program on DigitalBrain Atlasing
2 Task Forces – Neuroimaging (NIDASH) – Electrophysiology
NIDM working group
• NIDASH Task force – “Standards for Data Sharing aims to develop
generic standards and tools to facilitate the recording, sharing, and reporting of neuroscience metadata, in order to improve practices for the archiving and sharing of neuroscience data.”
• BIRN Derived Data Working Group
6
From XCEDE to NIDM
• XML-Based Clinical Experiment Data Exchange Schema (XCEDE): www.xcede.org – Describes subject, study, activation – Limited provenance encoding – Initiative of the BIRN
• XCEDE-DM • NeuroImaging Data Model (NIDM):
www.nidm.nidash.org
7
NIDM: Neuroimaging Data Model
8
Source: Poline et al, Frontiers in Neuroinformatics (2012).
NIDM: Neuroimaging Data Model
• Based on PROV-DM • First applications
– Description of the dataset-experiment hierarchy – Freesurfer volumes – DICOM terms
9
Meta-analysis use case CONTEXT
10
Results
Meta-analysis
Results
Results
Why meta-analyses?
• Increase statistical power • Combine information across studies
Data acquisition Analysis
Experiment Raw data Results
Data acquisition Analysis
Experiment Raw data Results
… Results
Meta-analysis
11
Data analysis in neuroimaging Analysis
Results Experiment
Data acquisition
Raw data Paper
Publication
MRI acquisition parameters
Task design and timing
Description of
participants
Mental processes
studied
Imaging data
12
500MB/subject 20GB
2.5GB/subject 100GB
[ ~2GB for stats]
< 0.5MB 0MB
Data processing and analysis procedure
Meta data
Data analysis in neuroimaging
Table of local maxima (quantitative)
13
Paper
Publication
?
Detection images (qualitative)
Peaks (quantitative)
< 0.5MB
Coordinate- or Image-Based meta-analysis?
14
Data acquisition Analysis
Experiment Raw data Results
Data acquisition Analysis
Experiment Raw data Results
…
Publication
Publication
Paper
Paper
Coordinate-based meta-analysis
Image-based meta-analysis
Shared results Data sharing
Data sharing environment CONTEXT
15
Acquisition
Pre-processing
Statistical analysis
Data sharing tools
Publication
16
Three major software packages
17
Automatically created with Neurotrends based on over 16 000 journal articles; Source: http://neurotrends.herokuapp.com/static/img/temporal/pkg-prop-year.png
~80%
Summary of the problem
• Use-case: Support meta-analysis • Machine-readable format describing
neuroimaging results
• Easiness for the end-user • Integrate with existing neuroimaging software
packages (SPM, FSL, AFNI,…) • Extend previous work: NIDM
18
NIDM FOR META-ANALYSIS
19
NIDM-Results NIDM FOR META-ANALYSIS
20
Neuroimaging Data Model
21
NIDM-Results
22
23
NIDM-Results
24
NIDM-Results: SPM-specific
25
NIDM-Results: FSL-specific
Standardization across software
• Model of the error – Prob. distribution: – Variance:
– Dependence:
26
heterogeneous
Independent noise
homogeneous
Compound Symmetry
Serially correlated
Arbitrarily correlated
Gaussian Non-‐Parametric …
global
local
regularized
Error models : SPM, FSL and AFNI
27
1st level 2nd level
Gaussian Serial. corr. global
Homogeneous local
Gaussian Independent noise
Homogeneous local
Gaussian Homogeneous local
Serial. corr. regularized
Gaussian Homogeneous local
Serial. corr. local
Gaussian Independent noise
Heterogeneous local
Gaussian Independent noise
Hetero-‐ or Homogeneous
local
Error models: non-parametric
28
2nd level: Sign-flipping
Heterogeneous local
Independent noise
NonParametric Symmetric
2nd level: Label permutation
Homogeneous local
Exchangeable noise local
NonParametric
Terms
• “Flat” ontology • Terms re-use:
– Interaction with STATO (statistical terms) – Dublin Core (file formats) – But also: NCIT, OBI…
• Work-in-progress • http://tinyurl.com/nidm-results/terms
• Aim: include the terms in Neurolex.
29
Queries • For each contrast get name, contrast file,
statistic file and type of statistic used.
30
prefix prov: <http://www.w3.org/ns/prov#> prefix nidm: <http://www.incf.org/ns/nidash/nidm#> SELECT ?contrastName ?contrastFile ?statType ?statFile WHERE { ?cid a nidm:ContrastMap ; nidm:contrastName ?contrastName ; prov:atLocation ?contrastFile . ?cea a nidm:ContrastEstimation . ?cid prov:wasGeneratedBy ?cea . ?sid a nidm:StatisticMap ; nidm:statisticType ?statType ; prov:atLocation ?statFile . }
More queries: http://tinyurl.com/nidm-results/query
Implementation NIDM FOR META-ANALYSIS
31
Implementation
• NIDM export – SPM12 (natively) – Scripts for FSL:
https://github.com/incf-nidash/nidm-results_fsl – First contact with AFNI developers
32
Future directions NIDM FOR META-ANALYSIS
33
Next steps and future plans
• Extend NIDM-Results implementation: – AFNI – SnPM, Randomise
• Refine the terms and definitions.
34
Next steps and future plans • NIDM import for
Neurovault
35
NIDM-Results and NIDM-Workflow
• NIDM-Results: effort on standardization across software – A small number of generic activities – A few software-specific entities
• NIDM-Workflow: a detailed view of all software-specific processes.
36
CONCLUSION
37
Conclusion
• NIDM-Results: standardized reporting of neuroimaging results – Use-case: Meta-analysis – Terms: http://tinyurl.com/nidm-results/terms/ – Specification: http://nidm.nidash.org – Implementation in SPM12 & FSL
• Next steps – Refine the terms, AFNI and SnPM/Randomise models – Integration with Neurovault – Build apps
38
Resources • Github: https://github.com/incf-nidash • Specifications: http://nidm.nidash.org
39
Acknowledgements
40
Thank you! To all the INCF NIDASH task force members.
NIDM working group Tibor Auer, Gully Burns, Fariba Fana, Guillaume Flandin, Satrajit Ghosh, Chris Gorgolewski, Karl Helmer, David Keator, Camille Maumet, Nolan Nichols, Thomas Nichols, Jean-Baptiste Poline, Jason Steffener, Jessica Turner.
INCF NIDASH - Other members David Kennedy, Cameron Craddock, Stephan Gerhard, Yaroslav Halchenko, Michael Hanke, Christian Haselgrove, Arno Klein, Daniel Marcus, Franck Michel, Simon Milton, Russell Poldrack, Rich Stoner.
This work is supported by the
41
Q & A
• Github: https://github.com/incf-nidash • Specifications: http://nidm.nidash.org
NIDM Resources