NIDM-Results: a Neuroimaging Data Model to share brain mapping statistical results

Only a tiny fraction of the data and metadata produced by an fMRI study is finally conveyed to the community. This lack of transparency not only hinders the reproducibility of neuroimaging results but also impairs future meta-analyses. In this work we introduce NIDM-Results, a standard providing a machine-readable description of neuroimaging statistical results along with key image data summarising the experiment. NIDM-Results provides a unified representation of mass univariate analyses including a level of detail consistent with available best practices. This standard allows authors to relay methods and results in a standard format that is not tied to a particular neuroimaging software package. Tools are available to export NIDM-Result graphs and associated files from the widely used SPM and FSL software packages, and the NeuroVault repository can import NIDM-Results archives. The specification is publically available at: http://nidm.nidash.org/specs/nidm-results.html.


Introduction
A neuroimaging technique like functional Magnetic Resonance Imaging (fMRI) generates hundreds of gigabytes of data, yet only a tiny fraction of that information is finally conveyed to the community. In a typical paper, the entire results report consists of 1) a list of significant local maxima, i.e. locations in the brain defined in a standard atlas space inferred to be distinct from noise, and 2) a graphical representation of the activations as an image figure.
This practice is unsatisfactory for three reasons. First, because it represents a massive loss of information from the raw and even the derived data used to draw the conclusion of the study. For example, a metaanalysis (in settings other than neuroimaging) combines estimates of effects of interest and their uncertainty across studies. In brain imaging, the locations of local maxima have no measures of uncertainty reported. While neuroimaging meta-analysis methods for coordinate data exist [1][2][3] they are a poor approximation to the meta-analysis that would be obtained if the image data were available 4 . Even though there are emerging infrastructures to support sharing of neuroimaging data (e.g. NeuroVault RRID:SCR_003806 5,6 ), these are still rarely utilised.
Second, despite the availability of guidelines [7][8][9] , ambiguous or incomplete methodological reporting in papers is still commonplace 10 hindering the robustness and reproducibility of scientific results 10,11 .
Finally, key methodological details of the study are described in free-form text in a paper and not available in machine-readable form, making these metadata essentially unsearchable. Databases have been built to provide metadata associated with published papers, either manually curated (e.g. BrainMap 12,13 ) or automatically-populated using text-mining algorithms (e.g. NeuroSynth 14,15 ), but, ideally, these metadata should be made available by the authors themselves at the time of the publication, together with the data. Additionally, searchable metadata, could help identify potential confounding factors that are currently being overlooked (e.g. how different smoothing kernels impact the meta-analysis, or the influence of different processing strategies on the outcome of the analysis).
In order to make neuroimaging results available in a machine-readable form a number of key technical issues have to be addressed. First, the scope of the metadata to be shared must be defined. The space of possible metadata to report is extremely large encompassing experimental design, acquisition, preprocessing, statistical analysis, etc. The optimal set of metadata is highly dependant on the application of interest and possible applications of shared data are broad. For example, in a meta-analysis, the contrast standard error map is required, while a comparison across neuroimaging processing pipelines would require a complete description of the analysis pipeline including software-specific parameterization.
Another technical issue is the need to define a common representation across neuroimaging software packages. While the three main neuroimaging software packages, SPM (RRID:SCR_007037) 16,17 , FSL (RRID:SCR_002823) 18, 19 and AFNI (RRID:SCR_005927) 20,21 , all implement similar analyses, they often use different terms to refer to the same concept. For example, FSL's parameter estimate maps (e.g. pe1.nii.gz) are the equivalent of SPM's beta maps (e.g. beta_0001.nii). They can also use the same term when referring to different concepts. For example, SPM and FSL both use a global scaling of the data to get "percent BOLD signal change", but due to differences in how the brain mask and mean signal are computed, the data are scaled quite differently 22 and are not comparable. In order to fully describe an analysis, the sharing of software-specific batch scripts (e.g. SPM matlabbatch files, FSL fsf files, or history stored in AFNI brick headers) would be a simple solution to provide all the parameters from an analysis, but the ability to compare and query across software would still be lacking. Pipeline systems like NiPype 23 , LONI Pipeline 24 and CBRAIN 25 do explicitly model analysis steps, but a large volume of research is still conducted directly with tools not embedded in pipelines. Ideally, one should be able to identify all studies corresponding to a set of criteria of interest regardless of the software used. This will only be possible if information about results across software can be represented using common data elements and structures.
A machine-readable representation of neuroimaging data and results, using a common descriptive standard across neuroimaging software packages, would address these issues of comparability and transparency.
A previous effort in this direction was the XML-based Clinical and Experimental Data Exchange (XCEDE) schema 26 , developed in the context of the Biomedical Informatics Research Network (BIRN) 27,28 . XCEDE modelled information about both the experimental design and results (peaks and clusters) in neuroimaging studies. This XML schema was defined to be independent from any particular neuroimaging analysis software and was made openly available 29 . XCEDE has been used by multiple sites across the United States and the United Kingdom in the context of the fBIRN project and is still in . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The core motivation of NIDM is to support data sharing and data reuse in neuroimaging by providing rich machine-readable metadata. Since its first developments in 2011, NIDM has been an ongoing effort and is currently comprised of three complementary projects: NIDM-Experiment, NIDM-Workflows and NIDM-Results. NIDM-Experiment targets the representation of raw data generated by the scanner and information on the participants. NIDM-Workflows focuses on the description of data analysis parameterization, including detailed software-specific variations. NIDM-Results, presented here, deals with the representation of mass-univariate neuroimaging results using a common descriptive standard across neuroimaging software packages.
A motivating use case for NIDM-Results was neuroimaging meta-analysis, but the format also produces a detailed machine-readable report of many facets of an analysis. The implementation of NIDM-Results within SPM and FSL, two of the main neuroimaging software packages, provides an automated solution to share maps generated by neuroimaging studies along with their metadata. While NIDM-Results focuses on mass-univariate studies and is mostly targeted at fMRI, the standard is also suitable for . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under Neuroimaging data sharing Task Force (NIDASH) which comprises a core group of experts representing more than ten labs involved in various facets of neuroimaging (including statistical analysis, informatics, software development, ontologies). It also involved close collaboration with the main neuroimaging software developers. The format is natively implemented in SPM and a NIDM-Results exporter is available for FSL and will be integrated in a future version of FSL. Both NeuroVault and CBRAIN support export to NIDM-Results and NeuroVault additionally can import NIDM-Results archives.

Definitions
The definitions provided below are used throughout the manuscript: • NIDM-Results graph: A particular instance of a representation of data and metadata complying with the NIDM-Results specification.
• NIDM-Results serialization: A text file rendering of a NIDM-Results graph.
• NIDM-Results pack: A compressed file containing a NIDM-Results serialization and some or all of the referenced image data files.

Overview
The NIDM-Results standard is defined by a W3-style specification, publicly available at http://nidm.nidash.org/specs/nidm-results.html and by an ontology (owl) file available at http://bioportal.bioontology.org/ontologies/NIDM-RESULTS. It is comprised of a controlled vocabulary, as well as instructions of how to use PROV to represent mass-univariate neuroimaging results. The model provides terms to describe key elements of neuroimaging methods using a common framework across neuroimaging software packages. For example, as illustrated in Fig. 1, error models are described in terms of assumed variance (homoscedastic, heteroscedastic) and assumed covariance structure . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint (independent, spatially correlated, etc.) and how these structures vary in space (defined independently at each voxel, globally throughout the brain or spatially regularised).
The current version, NIDM-Results 1.3.0, defines 214 terms (140 classes and 74 attributes) of which 45 terms are re-used from external vocabularies and ontologies. All terms are defined as specialisations of the PROV terms. Three namespaces are defined: http://purl.org/nidash/nidm# ("nidm:"), http://purl.org/nidash/spm# ("spm:") and http://purl.org/nidash/fsl# ("fsl:"). Anything that could be represented across software or that is a generic concept is defined in the "nidm:" namespace. Softwarespecific namespaces: ("spm:", "fsl:") are reserved for the description of functionality unique to one software (e.g. global null inference for conjunction testing in SPM). • 26 types of entities (of which 6 are optional) representing inputs and outputs of the activities; • 3 agents representing the 'neuroimaging analysis software', the 'person' or 'study group population' who participated in the study and the type of 'imaging instrument' used.
The statistical model is described in the 'design matrix' and 'error model' entities that are both used by the 'model parameter estimation' activity. The 'data' entity describes the scaling applied to the data before model fitting (especially relevant for first-level fMRI experiments) and links to the participants (as a 'person' or a group) and the 'imaging instrument' used to acquire the data (e.g. a magnetic resonance imaging scanner or an electroencephalography machine). A set of 'parameter estimate maps' is generated by the 'model parameter estimation' activity along with the analysis 'mask map', a 'residual mean squares map' and a 'grand mean map' that can be used to check the performance of the data . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint scaling. Optionally, a 'resels per voxel map' can also be generated to record local variations in noise smoothness. The 'contrast estimation' activity uses a subset of the 'parameter estimate maps', the 'residual mean squares map' and the analysis 'mask map' and combine them according to a 'contrast weight matrix' to generate a 'statistic map'. For T-tests, a 'contrast map' along with its 'contrast standard error map' are also generated while for F-tests a 'contrast explained mean square map' (i.e. the numerator of an Fstatistic) is provided.
Finally, the 'inference' activity uses a 'statistic map' and generates an 'excursion set map' given a 'height threshold' and an 'extent threshold'. The 'peak definition criteria' and 'cluster definition criteria' entities, used by 'inference', provide the connectivity criterion and minimal distance between peaks (e.g. default is set to 8 mm for SPM and 0 mm for FSL). The 'inference' activity can be replaced by a 'conjunction inference' which uses more than one statistic map. An optional 'display mask map' entity can be used to represent contrast masking, i.e. to restrict the display without affecting the correction for multiple comparisons. The 'inference' activity also generates the 'search space mask map' that represents the search region in which the inference was performed (i.e. the intersection of all input mask maps, except for the display mask map). A set of 'supra-threshold clusters' is derived from the 'excursion set map' and a set of 'peaks' is derived from each cluster. Those are the clusters and peaks that are typically reported in the results of a neuroimaging study.
A 'neuroimaging analysis software' agent represents the software package used to compute the analysis.
This agent is associated with all activities within the bundle.
Provenance of the 'NIDM-Results bundle' is also recorded: the bundle was generated by a 'NIDM-Results Export' activity which was performed by a 'NIDM-Results Exporter' software agent corresponding to the software used to create the NIDM-Results document (e.g. FSL's Python scripts, named 'nidmfsl' or SPM's exporter named 'spm_results_nidm'). The bundle is associated with a version number corresponding to the version of NIDM-Results model in use.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint Each activity, entity and agent has a number of predefined attributes. For instance, the list of attributes of an 'error model' entity is provided in Fig. 1.

Updates
Each release of NIDM-Results is associated with a version number. Comments on the current version as well as suggestions of extension can be provided on the GitHub nidm repository: https://github.com/incfnidash/nidm. Each extension or proposition of update will be reviewed and discussed with the members of the INCF NIDASH task force. Implementation SPM12 natively supports export of its results into a NIDM-Results pack, either by the use of a contextual menu in the results table or non-interactively via the batch interface as illustrated in Fig. 3. Export of FEAT results from FSL into a NIDM-results pack can be performed using the Python module nidmfsl (https://pypi.python.org/pypi/nidmfsl), as also illustrated in Fig. 3. nidmfsl was integrated in NeuroVault and as a plugin 38 of the CBRAIN web platform for high-performance computing (RRID:SCR_005513) 25 .
As a result, any FSL FEAT analysis uploaded to NeuroVault or performed in CBRAIN can be exported as a NIDM-Results pack. NeuroVault also accepts NIDM-Results packs as a mean to upload new data to a collection. unrevisioned, ds000107 unrevisioned, ds000114 unrevisioned).

Meta-analysis
From 21 pain studies (10 analysed in SPM and 11 in FSL) represented in NIDM-Results we performed group coordinate-based and image-based meta-analyses contrasting the effect of pain. The data and . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint Python script used to perform these meta-analyses are available on NeuroVault (http://neurovault.org/collections/1425/) and GitHub 41 respectively. The coordinate-based meta-analysis was performed using a Multilevel Kernel Density Analysis (MKDA) 2 .
Each NIDM-Results pack was queried to retrieve the coordinates of the local maxima, the reference space in use and the number of subjects for each contrast. Areas of significant activation (p<0.05 FWE cluster-wise with a cluster-forming threshold of p<0.001 uncorrected) found by the pain meta-analysis are displayed in Fig. 6.
In line with previous results from the literature 4 , the detections for the coordinate-based and image-based meta-analysis appear consistent with a lower sensibility of the coordinate-based meta-analysis.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint Table 2 provides a mapping between the guidelines provided in 7 to report neuroimaging results and the fields available in a NIDM-Results serialisation. NIDM-Results cover all elements from the "Statistical modelling" checklist that could be automatically retrieved within the neuroimaging software package. Examples of reports generated from a NIDM-Results export of group and single-subject analyses performed in SPM and FSL are provided in Fig. 7. The data and Python script used to generate those report are available on NeuroVault (http://neurovault.org/collections/1435/) and GitHub 41 respectively.

Discussion
Data sharing in the neuroimaging community is still restrained by a number of psychological and ethical factors that are beyond the scope of the current paper (see 42,43 for a review). Those will have to be addressed in order for data sharing to become common practice in the neuroimaging community. In an effort to address the technological barriers that make data sharing challenging, here we have proposed a solution to share neuroimaging results of mass univariate analyses.
As a first step to provide machine-readable metadata, we restricted our scope to information that was automatically extractable and attributes that were crucial for meta-analysis (e.g. number of subjects). This . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint NIDM-Results currently focuses on the representation of parametric mass-univariate analyses. Thanks to the intrinsic extensibility of RDF models, variants could be proposed to broaden its scope. For example, an extension for non-parametric statistics is under discussion 48 . Mass-univariate results, as the most well established approach for fMRI analysis, was an obvious choice to start a cross-software modelling effort. But neuroimaging cannot be limited to mass-univariate analyses and future work will focus on providing extensions for other types of analyses (e.g. analysis of resting state fMRI).
We based our modelling effort on PROV, a specification endorsed by the W3C, to model provenance on the web. Other efforts have been proposed to model provenance including families of ontologies like the OBO foundry 49 or DOLCE 50 . We chose PROV as it is lightweight, focused only on provenance, and is easily extensible to provide domain-specific knowledge.
Another recent effort to provide structured organisation of neuroimaging data is the Brain Imaging Data Structure (BIDS) 51 . While NIDM-Results and BIDS both concern the organisation and description of neuroimaging data, they operate at very distinct domains of the analysis path. BIDS provides a mechanism for organising only the original raw data, and it does not cover any derived data nor the definition of particular statistical models or the outputs of those models. NIDM-Results, in contrast, works at the other end of the analysis pipeline, defining a framework to describe the statistical model, the statistical 'contrasts' that interrogate the model, and the resulting statistical maps and inferences obtained from each contrast. Whereas BIDS was designed so that an end-user could manually create the files and directories of a BIDS-compliant data structure, NIDM-Results is intended to be automatically generated from analysis software and was therefore created using more expressive semantic web technologies.
NIDM-Results is based on RDF and semantic web technologies. While a number of ontologies have been developed in relation with neuroimaging (e.g. Cognitive Atlas 44 , CogPO 46 , OntoNeurolog 52 ), the use of controlled vocabularies and of linked data is not yet common practice in our community. As more and more data become available online and as standardisation effort like the RII develops, we believe that these technologies will become more widespread. RDF was chosen as a basis for NIDM for the expressivity of its graph-based structures, the possibility to form intricate queries across datasets 36 , as . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint well as for the extensibility of the created data models and the possibility to interconnect across knowledge domains (cf. 53 for a review).
One limitation of NIDM-Results is that only limited provenance is represented. For instance, computational environment, which has been shown to be source of undesired variability in neuroimaging results 54 , is not part of our model. NIDM-Results is part of a broader effort (NIDM) that aims at representing different levels of provenance in neuroimaging experiments. While those efforts are still under development, our goal is to keep a link between those components to eventually provide a complete representation of neuroimaging provenance.
As for the definition of any new model, gaining acceptance within the neuroimaging community will be crucial for NIDM-Results. To insure a level of consensus, including the point-of-view of different actors in the field, NIDM-Results was built as part of a collaborative effort. More feedback from the community is welcome and can be submitted as issues in our GitHub repository or by email at nidm-users@googlegroups.com. We also made a strong commitment to make implementations available. We believe NIDM-Results is an essential tool for the future of transparent, reproducible science using neuroimaging. If all research publications were accompanied by such a machine-readable description of . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint the experiment, debates on the exact methodology used would be compressed or eliminated, and any replication efforts greatly facilitated.

Process
Since August 2013 the model was developed through weekly teleconferences and eight focused workshops during which the team of experts iteratively defined the terminology, seeking to ensure that the output of AFNI, FSL and SPM could be represented in this framework. Furthermore, a separate meeting was organised with each of the development teams of SPM, FSL and AFNI to discuss the model and its included it in the model and depend on the exporters to generate it from existing data.

Term re-use and definitions
For each piece of information, we checked if an appropriate term was available in publicly available ontologies: in particular STATO for statistics term, PROV for provenance, NeuroLex for neuroscience terms, RRID for tools and also, to a lesser extent, Dublin Core, the NEPOMUK file ontology and the . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint Cryptographic Hash Functions vocabulary. Namespaces of the re-used ontologies are provided in Table 2. More details on the re-used vocabularies are provided below.

PROV
The W3C specification PROV defines three types of objects: an Activity represents a process that was performed on some data (e.g. a voxel-wise inference) and occurred over a fixed period of time; an Agent represents someone (human, organization, machine…) that takes responsibility for an activity (e.g. the SPM software) and, finally, an Entity represents any sort of data, parameters etc. that can be input or output of an activity (e.g. a NIfTI image). PROV also defines a set of relations between those objects (e.g. a voxel-wise inference Activity used a NIfTI image Entity; a voxel-wise inference Activity was associated with the SPM Agent and another NIfTI image Entity was generated by the segmentation Activity). NIDM-Results terms were defined as specialisations of PROV terms.

STATO
GLM analyses of fMRI data rely on well-known statistical constructs (e.g. one-sample T-test, two-sample T-test, F-tests, ANOVA, inference, ordinary least squares estimation, etc.). The general-purpose STATistics Ontology (STATO) is built on the top of the OBO foundry and aims to provide a set of terms describing statistics. We re-used statistics terms available in STATO (e.g. obo:'t-statistic') and when we could not find an appropriate statistical term, we engaged with STATO developers through GitHub issues to propose new terms (e.g. "residual mean squares" discussed in issue 35 58 ).

NeuroLex and RRID
Much work has been done in the neuroimaging community to provide controlled vocabularies and ontologies defining neuroimaging concepts. NeuroLex 59,60 provides a common platform that gathers terms from different sources (including previous vocabularies developed by NIF, BIRN…). Interestingly, Neurolex was part of the recent Resource Identification Initiative (RII) 61,62 that publicized the use of those identifiers (e.g. "RRID:SCR_007037" for SPM 63 ) in research papers. RII is currently focused on the identification of biological resources and has been quickly adopted, with more than 100 journals participating to date. We re-used the available RRIDs describing neuroimaging software packages.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint

Dublin Core, NEPOMUK file ontology and the Cryptographic Hash Functions vocabulary
Many vocabularies and ontologies have terms available to describe files. We chose to rely on the widely adopted DUBLIN core terminology. Additionally, we used the "fileName" term from the NEPOMUK file ontology and the SHA-256 term from the Cryptographic Hash Functions vocabulary.

New terms
When no term was found to describe a given neuroimaging concept of interest, we created a new term and carefully crafted a definition or engaged with the relevant ontology maintainers (e.g. STATO) to propose a new definition. All new terms and definitions were thoroughly discussed between our panel of experts in the NIDM working group, which is part of the INCF Neuroimaging Task Force (NIDASH).

Meta-analysis
Data collection was subject to the Oxford University ethics review boards, who approved the experiments.
Only statistical summaries with no identifying data are shared along this manuscript.
Results from 21 pain studies previously analysed with FSL were made available to us. The second-level analyses were recomputed with SPM for 10 of those studies in order to obtain a dataset of NIDM-Results packs coming from mixed software packages. We computed a one-sample meta-analysis contrasting the effect of "pain" and compared the results of coordinate-based and image-based meta-analyses.
The MKDA toolbox 64 , was used to perform the coordinate-based meta-analysis. The nidmresults Python toolbox (https://pypi.python.org/pypi/nidmresults) was used to generate the csv file required as input for the analyses.
FSL's FLAME 1 65 was used to compute the image-based meta-analysis with the gold standard approach (3rd level mixed-effects general linear model). Difference in data scaling between software packages were compensated by rescaling the FSL maps to a target intensity of 100 (instead of 10 000 by default).
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint

Reporting of neuroimaging results
From four studies exported with NIDM-Results we wrote a script to extract the information of interest to describe group and subject-level statistics using the RDFlib library 66 to query the documents. The paragraph that was generated could, for instance, be used as part of the method section in a research       . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint  Fig. 4. Image-based and coordinate-based meta-analyses using NIDM-Results. Each NIDM-Results pack is queried to retrieve the data and metadata of interest for each type of meta-analysis. These data are then combined in a meta-analysis. . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint prefix prov: <http://www.w3.org/ns/prov#> prefix nidm: <http://purl.org/nidash/nidm#> prefix contrast_estimation: <http://purl.org/nidash/nidm#NIDM_0000001> prefix contrast_map: <http://purl.org/nidash/nidm#NIDM_0000002> prefix stderr_map: <http://purl.org/nidash/nidm#NIDM_0000013> prefix contrast_name: <http://purl.org/nidash/nidm#NIDM_0000085> prefix statistic_map: <http://purl.org/nidash/nidm#NIDM_0000076> prefix mask_map: <http://purl.org/nidash/nidm#NIDM_0000054>  . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted July 18, 2016. ; https://doi.org/10.1101/041798 doi: bioRxiv preprint   . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under

Location of the search space image
Attribute prov:atLocation of a 'Search Space Mask Map' entity.
Volume of the search region in voxels.