ANIMA: A data-sharing initiative for neuroimaging meta-analyses

Andrew T. Reid, Danilo Bzdok, Sarah Genon, Robert Langner, Veronika I. Müller, Claudia R. Eickhoff, Felix Hoffstaedter, Edna-Clarisse Cieslik, Peter T. Fox, Angela R. Laird, Katrin Amunts, Svenja Caspers (+1 others)
2016 NeuroImage  
1 2 Available online xxxx 16 17 Meta-analytic techniques allow cognitive neuroscientists to pool large amounts of data across many individual 18 task-based functional neuroimaging experiments. These methods have been aided by the introduction of online 19 databases such as Brainmap.org or Neurosynth.org, which collate peak activation coordinates obtained from 20 thousands of published studies. Findings from meta-analytic studies typically include brain regions which are 21 consistently
more » ... across studies for specific contrasts, investigating cognitive or clinical hypotheses. These 22 regions can be subsequently used as the basis for seed-based connectivity analysis, or formally compared to neu-23 roimaging data in order to help interpret new findings. To facilitate such approaches, we have developed a new 24 online repository of meta-analytic neuroimaging results, named the Archive of Neuroimaging Meta-analyses 25 (ANIMA). The ANIMA platform consists of an intuitive online interface for querying, downloading, and contrib-26 uting data from published meta-analytic studies. Additionally, to aid the process of organizing, visualizing, and 27 working with these data, we present an open-source desktop application called Volume Viewer. Volume Viewer 28 allows users to easily arrange imaging data into composite stacks, and save these sessions as individual files, 29 which can also be uploaded to the ANIMA database. The application also allows users to perform basic functions, 30 such as computing conjunctions between images, or extracting regions-of-interest or peak coordinates for fur-31 ther analysis. The introduction of this new resource will enhance the ability of researchers to both share their 32 findings and incorporate existing meta-analytic results into their own research. 33 Brainmap.org (Laird et al., 2011), Neurovault.org (Gorgolewski 43 et al., 2015), and Neurosynth.org (Yarkoni et al., 2011) provide ac-44 cess to data from hundreds of published task-based fMRI studies, in 45 standard coordinates (recently reviewed by Fox et al., 2014). In 46 addition to extensive meta-analyses based upon manual search tech-47 niques, these platforms have facilitated a growing number of meta-48 analytic studies of the neural correlates of specific cognitive functions, 49 using methods such as multilevel kernel density analysis (MKDA; 50 Wager et al., 2007) and activation likelihood estimation (ALE; Eickhoff 51 et al., 2009, 2012). Meta-analysis entails the pooling of data over tens 52 to thousands of individual studies, and thus provides: (1.) greater 53 sensitivity and specificity to detect "true" effects; (2.) a means of deter-54 mining core groups of brain regions subserving a specific task or charac-55 terizing a specific disease; and (3.) a method for formal comparison of 56 different subfacets of a cognitive domain. This approach has been 57 used, for instance, to demonstrate the neural correlates of sustained at-58 tention (Langner and Eickhoff, 2013), investigate face processing areas 59 in autistic subjects (Nickl-Jockschat et al., 2014), and identify key re-60 gions subserving supervisory attentional control (Cieslik et al., 2015). 61 Results from meta-analyses can be subsequently used as robust prior 62 information in the design of task-based fMRI studies, and as regions-of-63 interest (ROIs) for connectivity methods based on functional correla-64 tions (e.g., Müller et al., 2014; Schilbach et al., 2014), or virtually any 65 other type of seed-based analysis. This includes topical meta-analytic 66 approaches such as ALE and MKDA, as well as methods which use 67 meta-analysis to infer functional connectivity, such as meta-analytic 68 connectivity modelling (MACM; Etkin and Wager, 2007; Kober et al., 69 2008; Robinson et al., 2010), in which functional coactivations are NeuroImage xxx (2015) F 70 assessed across all tasks in the database (see also Xue et al., 2014). Ad-71 ditional meta-analytic approaches include the parcellation of the brain 72 into functionally distinct subregions, such as coactivation-based 73 parcellation (CBP; Chang et al., 2012; Eickhoff et al., 2011; Northoff 74 et al., 2006). A typical ALE analysis, for example, will result in a distinct 75 ROI or set of ROIs that are associated with a particular psychological or 76 clinical feature (see Box 1). These results are then commonly used as 77 seed regions for further analysis of these features (e.g., zu Eulenburg 78 et al., 2012). Given their utility, results from meta-analytic studies are 79 becoming increasingly popular as starting points for future analyses, 80 and are hence commonly requested from the authors. However, this 81 mode of data exchange typically requires a time delay for locating, orga-82 nizing, packaging, and sending data, and can be complicated further by 83 confusion over the way data are named, the type of information they 84 represent, and data formats in which they are stored. Moreover, use of 85 these data in published articles would benefit from the ability to refer-86 ence a specific and permanent online location, as well as provenance 87 tracking, particularly for purposes of validation and replication. These 88 considerations present a need for a more standardized, easily accessible 89 means of sharing meta-analytic results, which has motivated the crea-90 tion of a new online data resource called the Archive of Neuroimaging 91 Meta-Analyses (ANIMA). This database can be accessed at http:// 92 anima.fz-juelich.de. 93 The concept behind the ANIMA database is simple: to provide the re-94 sults of published meta-analyses and coactivation-based parcellations 95 to interested parties, in the form of statistical maps or labels, encoded 96 as image files. This approach has a number of important requirements. 97 Firstly, data should be easily citable if reused for further analyses, and 98 thus full information about the source of the results must be provided 99 with them. In addition to metadata about the article itself (reference 100 details, cross-links to PubMed entries, abstract information, etc.), 101 there should be sufficient information provided to identify the data 102 represented by each image file, including its associated figure, if ap-103 plicable. Secondly, data retrieved from the database should be simple 104 to organize, visualize, and use. A user should be able to peruse the 105 web interface, query and select studies of interest, download these 106 to their local machine, and immediately browse and utilize the data 107 they have retrieved. Thirdly, in order for the database to expand 108 and provide a thorough sampling of the literature, it is important to 109 provide a convenient interface through which researchers can sub-110 mit data from their own studies. 111 With the ANIMA database, we have provided solutions to each of 112 these requirements. In what follows, we will describe our databasing 113 approach, which includes an intuitive online interface for querying, 114 downloading, and submitting data, as well as a stand-alone, cross-115 platform desktop tool for easily browsing, visualizing, and performing 116 common computations, such as obtaining a conjunction between im-117 ages. This new resource will provide researchers with a straightforward 118 means of including meta-analytic results in their studies, both as ROIs 119 for future analyses and as a point of comparison against new results. 120 The initial release of the database will include data from 25 published 121 meta-analytic studies, but is intended to grow in order to incorporate 122 the increasing number of studies being added to the literature. 123 Database overview 124 ANIMA is designed to serve a number of functions. Firstly, it is a 125 157 net/volumeviewer). The interface allows data to be organized within 158 a "library" framework, in which individual studies can be represent-159 ed. The data retrieved from ANIMA is already organized according to 160 this framework, so downloaded ANIMA studies can be immediately 161 imported and viewed in Volume Viewer. Furthermore, if a study in-162 cludes Volume Viewer session files, predefined sets of image com-163 posites, with custom colour mapping and template or atlas layers, 164 can be viewed with little effort on the part of the user. Importantly, 165 this tool also implements a number of utilities which allow users to B:1 Box 1 B:2 Common methodologies used in ANIMA studies. B:3 B:5 Three common approaches used in ANIMA studies are described B:6 below. B:7 Activation likelihood estimation (ALE): This is a meta-analytic B:8 approach through which peak activation foci, reported in stan-330 of this, additional support for ANIMA will be provided in terms of docu-331 mentation, and an online support forum which can be used to ensure 332 that common issues can be answered once, and a means of reporting 333 bugs, which can be of great assistance to the database developers. 334 The Volume Viewer tool is based on the ModelGUI API, which is an 335 open-source project hosted on http://www.launchpad.net/volume-336 viewer (an online platform designed to be scalable, and which supports 337 projects as large as the Ubuntu community). We intend to provide Vol-338 ume Viewer also as an open-source project, such that it can be freely ob-339 tained and developed by interested members of the neuroscience 340 community. A number of future improvements are planned for Volume 341 Viewer, including: the ability to specify named voxels and ROIs; the 342 ability to extract peaks from a smoothed image map; the addition of 343 3D volume and surface rendering; easy transfer of data between sur-344 faces and volumes; rendering of network graphs for the visualization 345 of connectivity information; and layout and printing features. 346 Summary 347 We present a new database which provides free and convenient on-348 line access to the results of published neuroimaging meta-analyses. 349 Data can be used for comparison with one's own results, or as a starting 350 point for new analyses, and the ANIMA interface provides a set of simple 351 tools which greatly facilitates this process. This interface includes a 352 search function, a form for submitting new studies, and an open-353 source stand-alone software tool for visualizing and organizing study 354 data, and generating ROIs for further analysis. It is our hope that the 355 ANIMA database will improve the way in which the results of meta-356 analytic neuroimaging studies will be used in the future, and encourage 357 researchers to incorporate these important results into their research 358 approaches. 359 References 360 Amft, M., Bzdok, D., Laird, A.R., Fox, P.T., Schilbach, L., Eickhoff, S.B., 2014. Definition and 361 characterization of an extended social-affective default network. Brain Struct. Funct. 362 http://dx.investigation of the struc-364 tural, connectional, and functional subspecialization in the human amygdala. Hum. 365 Brain Mapp. 34, 3247-3266. http://dx.2011. ALE meta-analysis on facial judgments of trustworthiness and attractive-368 ness. Brain Struct. Funct. 215, 209-223. http://dx.Characterization of the temporo-parietal junction by 372 combining data-driven parcellation, complementary connectivity analyses, and func-373 tional decoding. Neuroimage 81, 381-392. http://dx.
doi:10.1016/j.neuroimage.2015.07.060 pmid:26231246 fatcat:c2fpel4gp5gipfaxmhr7nfez3i