A spectrum of sharing: maximization of information content for brain imaging data
© Calhoun; licensee BioMed Central. 2015
Received: 27 October 2014
Accepted: 17 December 2014
Published: 29 January 2015
Efforts to expand sharing of neuroimaging data have been growing exponentially in recent years. There are several different types of data sharing which can be considered to fall along a spectrum, ranging from simpler and less informative to more complex and more informative. In this paper we consider this spectrum for three domains: data capture, data density, and data analysis. Here the focus is on the right end of the spectrum, that is, how to maximize the information content while addressing the challenges. A summary of associated challenges of and possible solutions is presented in this review and includes: 1) a discussion of tools to monitor quality of data as it is collected and encourage adoption of data mapping standards; 2) sharing of time-series data (not just summary maps or regions); and 3) the use of analytic approaches which maximize sharing potential as much as possible. Examples of existing solutions for each of these points, which we developed in our lab, are also discussed including the use of a comprehensive beginning-to-end neuroinformatics platform and the use of flexible analytic approaches, such as independent component analysis and multivariate classification approaches, such as deep learning.
KeywordsData sharing Privacy Classification Multivariate Neuroinformatics Deep learning Independent component analysis
Wide-spread sharing of neuroimaging data and results is gaining momentum despite initial bold attempts which failed to gain widespread adoption initially [1,2]. Recently, calls for neuroimaging data sharing have been revived , though there is a lack of consensus about ideal models for incentivizing data sharing. Indeed there are many issues to consider, such as when to best consider sharing (e.g., at study setup, or after study completion), incentives to both data providers and data consumers, resources and sustainability, type of data to be shared (e.g., summary results or raw data), as well as the use of analytic approaches ranging from a high-level summary (e.g., meta-analytic) to data-driven and multivariate approaches.
First, regarding data capture, it almost goes without saying that it makes sense to maximize the quality of data as it is collected. However, most studies still do not collect data fully electronically, do not store all aspects of study information in a central place, and still draw upon error prone data entry and use of individual spreadsheets. In addition, one of the most difficult aspects of sharing data across studies relates to mapping the assessment information into a common framework. Indeed, the development of common data elements for various domains is the focus of much research [4-6]. Secondly, the type of data shared ranges from higher to lower information density from time series data (e.g., an fMRI data set), to contrast or connectivity maps, to coordinates and peaks (e.g., tables in a journal article). And finally, the flexibility of the analytic approach is tightly tied into the availability of data at hand and also directly related to the amount of information one can extract from the data. We discuss all three of these issues and make some recommendations which we hope will be useful for the field.
In summary, the use of standardized tools for capturing and organizing data, is essential as they have been shown to both reduce errors, as well as increase efficiency of data capture [27-29]. There are many tools available for capturing assessment data [29-32], though such solutions are not used as much as they should be in neuroimaging studies, especially for neuroimaging data, and the studies that do tend to use separate systems for neuroimaging and assessment data. However there are some notable exceptions to this and a large growth in the number of neuroinformatics tools available to the community. The community will benefit greatly from an increase in integrated systems where querying for multiple data types (e.g., neuroimaging, assessment, genetics, social media) is possible via a single entry point.
It is obvious that if maximizing information is the only goal then sharing of raw data should always be done. However in some cases there are goals which compete against the maximization of information, such as the need to preserve privacy. In some cases privacy is of paramount importance and can be a major barrier to data sharing. High dimensional datasets entail a high risk for re-identification despite meeting current privacy standards (e.g., HIPAA) –a common concern in the context of high dimensional biological datasets (e.g., genetics, MRI images). The recent Netflix competition highlighted concerns about phenotypic data when some competitors inadvertently re-identified individuals from anonymous datasets  (http://www.netflixprize.com; http://www.wikipedia.org/wiki/Netflix_Prize). The well-known example of genetic reidentification from datasets anonymized per National Institutes of Health (NIH) guidelines is another cautionary tale [38-41].
Data usage agreements (DUA) are a potential solution for enabling access to data while maintaining participant privacy, but unfortunately they have significant limitations for large studies, for example getting approval for many DUAs, each of which may require institutional approach, can be cumbersome and slow. NIH’s centralized database efforts, such as the National Database for Autism Research (NDAR) , are a step forward, but are US-based and require a federal-wide assurance number (FWA), limiting the international sharing of data, and still requires centralized downloading and manual organization of all data. The incorporation of a DUA management tool is one possibility which would be extremely helpful for building large consortia.
Data sharing efforts like ADNI , HCP , INDI , and openfMRI  are open, provide deidentified data, and use the DUA approach. Other approaches, in particular ENIGMA , which work with more sensitive genetic data, do not require data to be shared, but instead work with individual investigators to have them run scripts on their data to provide desired summary measures for meta-analysis. This is more private, though not in a quantifiable way.
The development of privacy preserving analysis approaches is an example of maximizing information while addressing the important concern of privacy. The solution discussed also touches on the use of flexible analytic approaches, such as multivariate classification. Such tools are essential in our quest to make sense of the complex data we are collecting and ultimately, we hope, the human brain. Regarding sharing of raw (and preprocessed data), a recent large consortium (over 5,000 rest fMRI data sets) on reproducibility and replicability (CoRR) of resting fMRI is currently available through COINS and NITRC [52,53]. It will be very interesting to see how this data is used, and certainly it would be possible to systematically compare, and with larger numbers, the various points on the data sharing spectra that we discuss. Though sharing of raw data will always give the most flexibility, there are also great benefits to sharing intermediate data. For example, many interesting findings have emerged in the area of meta-analysis or of the analysis of statistical maps calculated from imaging data [45,54,55].
One key concern with the use of more complex analytic approaches is the potential for overfitting the data as well as the lack of interpretability, especially with nonlinear approaches. These are valid concerns, the first can be addressed by using best practices in cross-validation of results (e.g., k-fold cross-validation) and careful evaluation of potential confounding variables. The latter represents a desire to interpret the results. Approaches like ICA are quite often linear, and can thus be quite readily interpreted, and the most widely-used ICA approaches optimize for both independent and sparsity measures with considerable success . Fortunately, even for more complex methods, there are ways to project the data into a domain that can be interpreted. This however has not been a major goal of the initial wave of results, which primarily focus on classification performance. But even highly nonlinear approaches, such as deep learning, can be carefully evaluated at each of the layers to interpret the underlying results. However, much more work is needed in this area.
In summary, flexible data analysis approaches can be highly informative especially when the underlying signals of interest are complex and poorly understood. Ultimately, there is a trade-off in the use of a simpler model with fewer parameters; however, a simpler model does not guarantee a better solution. One example summarized in  shows, in the case of predicting age, the mean activity across the entire brain gives better predictive power over more complex approaches, yet when predicting diagnosis, a connectivity-based measure was more informative than the simpler measures. That being said, given the high complexity of the brain and questions we are asking, and the extremely simple models that are most widely used in brain imaging, there is substantial room for growth in the area of more flexible modeling approaches which will likely lead to an increased understanding of brain structure and function. This has already been born out, for example functional connectivity [34,64,65] which was initially dismissed by much of the field, has grown into a major research focus.
Data sharing in neuroimaging is alive and well. This review has focused upon the concept of maximization of information, which is extremely important if we are to move our understanding of the brain forward. Consider the fact that we are still finding new information within very complex fMRI data sets that was not initially revealed (such as the recent focus on time-varying connectivity ). Current approaches are taking a variety of practical shortcuts to push data sharing forward, such as focusing only on meta-analytic approaches or sharing of only contrast images. While such approaches have their place and are extremely useful, we must not lose sight of the goal of making all collected data available to the community. Within the domains of data capture, data density, and data analysis I have tried to provide some examples, challenges, and solutions in order to foster this ongoing discussion. I look forward to the future and believe the combination of 1) technological advances and tools to assist investigators in collection of high quality data in a way that can be easily shared; 2) approaches to confront storage and computational barriers associated with sharing of the most raw form of the data; and 3) advanced algorithms to enable data-mining of rich data sets even in the context of possible constraints, such as privacy concerns, will move the field ahead at a rapid pace to help fill in the huge gaps in knowledge we have about human brain function and ultimately may help improve the lives of those with devastating brain disease.
Collaborative informatics and neuroimaging suite
Data usage agreement
Functional magnetic resonance imaging
Functional network connectivity
Independent component analysis
National database for autism research
National Institutes of Health
The work was in part funded by NIH via a COBRE grant P20GM103472 and grants R01EB005846 and 1R01EB006841. The author would like to thank Alvaro Ulloa for help with the figures and Sergey Plis for helpful discussions.
- Van Horn JD, Grethe JS, Kostelec P, Woodward JB, Aslam JA, Rus D, et al. The Functional Magnetic Resonance Imaging Data Center (fMRIDC): the challenges and rewards of large-scale databasing of neuroimaging studies. Philos Trans R Soc Lond B Biol Sci. 2001;356:1323–39. 1088517.PubMed CentralPubMedView ArticleGoogle Scholar
- Van Horn JD, Gazzaniga MS. Opinion: Databasing fMRI studies towards a ‘discovery science’ of brain function. Nat Rev Neurosci. 2002;3:314–8.PubMedView ArticleGoogle Scholar
- Poline JB, Breeze JL, Ghosh S, Gorgolewski K, Halchenko YO, Hanke M, et al. Data sharing in neuroimaging research. Front Neuroinform. 2012;6:9. 3319918.PubMed CentralPubMedView ArticleGoogle Scholar
- Warner J, Johnston M, Korngut L, Jette N, Pringsheim T. Common data elements for neurological registries. Can J Neurol Sci. 2013;40:S62–3.PubMedView ArticleGoogle Scholar
- Grove MJ, Lamberty GJ, Gatewood LC, Johnson LM. Traumatic brain injury rehabilitation: analysis of common data elements. Stud Health Technol Inform. 2013;192:1186.PubMedGoogle Scholar
- Ghitza UE, Gore-Langton RE, Lindblad R, Shide D, Subramaniam G, Tai B. Common data elements for substance use disorders in electronic health records: the NIDA Clinical Trials Network experience. Addiction. 2013;108:3–8.PubMedView ArticleGoogle Scholar
- McGarvey PB, Ladwa S, Oberti M, Dragomir AD, Hedlund EK, Tanenbaum DM, et al. Informatics and data quality at collaborative multicenter Breast and Colon Cancer Family Registries. J Am Med Inform Assoc. 2012;19:e125–8. 3392863.PubMed CentralPubMedView ArticleGoogle Scholar
- Mennes M, Biswal BB, Castellanos FX, Milham MP. Making data sharing work: the FCP/INDI experience. Neuroimage. 2013;82:683–91.PubMed CentralPubMedView ArticleGoogle Scholar
- Obeid JS, McGraw CA, Minor BL, Conde JG, Pawluk R, Lin M, et al. Procurement of shared data instruments for Research Electronic Data Capture (REDCap). J Biomed Inform. 2013;46:259–65. 3600393.PubMed CentralPubMedView ArticleGoogle Scholar
- Collaborative informatics and neuroimaging suite (COINS). http://coins.mrn.org
- Scott A, Courtney W, Wood D, De la Garza R, Lane S, Wang R, et al. COINS: an innovative informatics and neuroimaging tool suite built for large heterogeneous datasets. Front Neuroinformatics. 2011;5:1–15. PMC3250631.View ArticleGoogle Scholar
- M King, D Wood, B Miller, R Kelly, W Courtney, D Landis, et al. “Automated collection of imaging and phenotypic data to centralized and distributed data repositories,”. Front Neuroinformatics. in press, PMC Journal - In Process.
- D Wood, M King, D Landis, W Courtney, R Wang, R Kelly, et al. “Harnessing modern web application technology to create intuitive and efficient data visualization and sharing tools”. Front Neuroinformatics. in press, PMC Journal - In Process.
- Loring DW, Lowenstein DH, Barbaro NM, Fureman BE, Odenkirchen J, Jacobs MP, et al. Common data elements in epilepsy research: development and implementation of the NINDS epilepsy CDE project. Epilepsia. 2011;52:1186–91. 3535455.PubMed CentralPubMedView ArticleGoogle Scholar
- Ugurbil K, Xu J, Auerbach EJ, Moeller S, Vu AT, Duarte-Carvajalino JM, et al. “Pushing spatial and temporal resolution for functional and diffusion MRI in the Human Connectome Project. Neuroimage. 2013;80:80–104. 3740184.PubMed CentralPubMedView ArticleGoogle Scholar
- Da Mota B, Tudoran R, Costan A, Varoquaux G, Brasche G, Conrod P, et al. Machine learning patterns for neuroimaging-genetic studies in the cloud. Front Neuroinform. 2014;8:31. 3986524.PubMed CentralPubMedView ArticleGoogle Scholar
- V Potluru, J Diaz-Montes, AD Sarwate, SM Plis, VD Calhoun, B Pearlmutter, et al. “CometCloudCare (C3). Distributed Machine Learning Platform-as-a-Service with Privacy Preservation,” in Neural Information Processing Systems (NIPS). Montreal, Canada, 2014.
- Milham MP. Open neuroscience solutions for the connectome-wide association era. Neuron. 2012;73:214–8.PubMedView ArticleGoogle Scholar
- Potkin SG, Turner JA, Brown GG, McCarthy G, Greve DN, Glover GH, et al. Working memory and DLPFC inefficiency in schizophrenia: The FBIRN study. Schizophr Bull. 2009;35:19–31.PubMed CentralPubMedView ArticleGoogle Scholar
- Van Horn JD, Toga AW. Multisite neuroimaging trials. Curr Opin Neurol. 2009;22:370–8. 2777976.PubMed CentralPubMedView ArticleGoogle Scholar
- Marcus DS, Harwell J, Olsen T, Hodge M, Glasser MF, Prior F, et al. Informatics and data mining tools and strategies for the human connectome project. Frontiers in Neuroinformatics. 2011;5:4. 3127103.PubMed CentralPubMedView ArticleGoogle Scholar
- Moeller S, Yacoub E, Olman CA, Auerbach E, Strupp J, Harel N, et al. Multiband multislice GE-EPI at 7 tesla, with 16-fold acceleration using partial parallel imaging with application to high spatial and temporal whole-brain fMRI. Magn Reson Med. 2010;63:1144–53. 2906244.PubMed CentralPubMedView ArticleGoogle Scholar
- Marcus DS, Olsen TR, Ramaratnam M, Buckner RL. The Extensible Neuroimaging Archive Toolkit: an informatics platform for managing, exploring, and sharing neuroimaging data. Neuroinformatics. 2007;5:11–34.PubMedGoogle Scholar
- Dinov I, Lozev K, Petrosyan P, Liu Z, Eggert P, Pierce J, et al. Neuroimaging study designs, computational analyses and data provenance using the LONI pipeline. PLoS One. 2010;5:2946935.View ArticleGoogle Scholar
- BIRN. http://www.birn-community.org
- Collaborative informatics and neuroimaging suite (COINS) Data Exchange. http://coins.mrn.org/dx
- Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, Conde JG. Research electronic data capture (REDCap)--a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform. 2009;42:377–81. 2700030.PubMed CentralPubMedView ArticleGoogle Scholar
- Mitchel JT, Kim YJ, Choi J, Park G, Cappi S, Horn D, et al. Evaluation of Data Entry Errors and Data Changes to an Electronic Data Capture Clinical Trial Database. Drug Inf J. 2011;45:421–30. 3777611.PubMed CentralPubMedView ArticleGoogle Scholar
- Fraccaro P, Dentone C, Fenoglio D, Giacomini M. Multicentre clinical trials’ data management: a hybrid solution to exploit the strengths of electronic data capture and electronic health records systems. Inform Health Soc Care. 2013;38:313–29.PubMedView ArticleGoogle Scholar
- Obeid JS, Gerken K, Madathil KC, Rugg D, Alstad CE, Fryar K, et al. Development of an electronic research permissions management system to enhance informed consents and capture research authorizations data. AMIA Jt Summits Transl Sci Proc. 2013;2013:189–93. 3845791.PubMed CentralPubMedGoogle Scholar
- King C, Hall J, Banda M, Beard J, Bird J, Kazembe P, et al. Electronic data capture in a rural African setting: evaluating experiences with different systems in Malawi. Glob Health Action. 2014;7:25878. 4216812.PubMedGoogle Scholar
- Dillon DG, Pirie F, Rice S, Pomilla C, Sandhu MS, Motala AA, et al. Open-source electronic data capture system offered increased accuracy and cost-effectiveness compared with paper methods in Africa. J Clin Epidemiol. 2014;67:1358–63.PubMed CentralPubMedView ArticleGoogle Scholar
- Calhoun VD, Allen E. Extracting Intrinsic Functional Networks with Feature-based Group Independent Component Analysis. Psychometrika. 2013;78:243–59. 2013, PMC Journal - In Process.PubMedView ArticleGoogle Scholar
- Calhoun VD, Miller R, Pearlson GD, Adalı T. The chronnectome: Time-varying connectivity networks as the next frontier in fMRI data discovery. Neuron. 2014;84:262–74. PMC Journal - In Process.PubMedView ArticleGoogle Scholar
- Sui J, Adalı T, Yu Q, Calhoun VD. A Review of Multivariate Methods for Multimodal Fusion of Brain Imaging Data. J Neurosci Methods. 2012;204:68–81. PMC3690333.PubMed CentralPubMedView ArticleGoogle Scholar
- Calhoun VD, Adalı T. Feature-based Fusion of Medical Imaging Data. IEEE Trans Inf Technol Biomed. 2009;13:1–10. PMC2737598.View ArticleGoogle Scholar
- http://www.netflixprize.com; http://www.wikipedia.org/wiki/Netflix_Prize.
- Homer N, Szelinger S, Redman M, Duggan D, Tembe W, Muehling J, et al. Resolving individuals contributing trace amounts of DNA to highly complex mixtures using high-density SNP genotyping microarrays. PLoS Genet. 2008;4:e1000167. 2516199.PubMed CentralPubMedView ArticleGoogle Scholar
- Cassa CA, Wieland SC, Mandl KD. Re-identification of home addresses from spatial locations anonymized by Gaussian skew. Int J Health Geogr. 2008;7:45. 2526988.PubMed CentralPubMedView ArticleGoogle Scholar
- El Emam K. Methods for the de-identification of electronic health records for genomic research. Genome Med. 2011;3:25. 3129641.PubMed CentralPubMedView ArticleGoogle Scholar
- Neamatullah I, Douglass MM, Lehman LW, Reisner A, Villarroel M, Long WJ, et al. Automated de-identification of free-text medical records. BMC Med Inform Decis Mak. 2008;8:32. 2526997.PubMed CentralPubMedView ArticleGoogle Scholar
- Hall D, Huerta MF, McAuliffe MJ, Farber GK. Sharing heterogeneous data: the national database for autism research. Neuroinformatics. 2012;10:331–9.PubMed CentralPubMedView ArticleGoogle Scholar
- Jack Jr CR, Bernstein MA, Fox NC, Thompson P, Alexander G, Harvey D, et al. The Alzheimer’s Disease Neuroimaging Initiative (ADNI): MRI methods. J Magn Reson Imaging. 2008;27:685–91. 2544629.PubMed CentralPubMedView ArticleGoogle Scholar
- Van Essen DC, Smith SM, Barch DM, Behrens TE, Yacoub E, Ugurbil K, et al. The WU-Minn Human Connectome Project: an overview. Neuroimage. 2013;80:62–79. 3724347.PubMed CentralPubMedView ArticleGoogle Scholar
- Poldrack RA, Barch DM, Mitchell JP, Wager TD, Wagner AD, Devlin JT, et al. Toward open sharing of task-based fMRI data: the OpenfMRI project. Front Neuroinform. 2013;7:12. 3703526.PubMed CentralPubMedView ArticleGoogle Scholar
- Thompson PM, Stein JL, Medland SE, Hibar DP, Vasquez AA, Renteria ME, et al. The ENIGMA Consortium: large-scale collaborative analyses of neuroimaging and genetic data. Brain Imaging Behav. 2014;8:153–82. PMC Journal - In Process.PubMed CentralPubMedGoogle Scholar
- AD Sarwate, SM Plis, J Turner, and VD Calhoun, “Sharing privacy-sensitive access to neuroimaging and genetics data: a review and preliminary validation”, Frontiers in Neuroinformatics. in press, PMC Journal - In Process.
- Sweeney L. k-anonymity: a model for protecting privacy. Int J Uncertainty Fuzziness Knowl Based Syst. 2002;10:557–70.View ArticleGoogle Scholar
- Chaudhuri K, Monteleoni C, Sarwate AD. Differentially private empirical risk minimization. J Mach Learn Res. 2011;12:1069–109.PubMed CentralPubMedGoogle Scholar
- ViPAR. http://www.the-scientist.com/?articles.view/articleNo/37622/title/Data-Drive/
- Murtagh MJ, Demir I, Jenkings KN, Wallace SE, Murtagh B, Boniol M, et al. Securing the data economy: translating privacy and enacting security in the development of DataSHIELD. Public Health Genomics. 2012;15:243–53.PubMedView ArticleGoogle Scholar
- XN Zuo, JS Anderson, P Belle, RM Birn, B Biswal, J Blautzik, et al. “An Open Science Resource for Establishing Reliability and Reproducibility in Functional Connectomics,” Scientific Data. in press, PMC Journal - In Process.
- Buccigrossi R, Ellisman M, Grethe J, Haselgrove C, Kennedy DN, Martone M, et al. The Neuroimaging Informatics Tools and Resources Clearinghouse (NITRC). AMIA Annu Symp Proc. 2008;6:1000.Google Scholar
- Laird AR, Eickhoff SB, Li K, Robin DA, Glahn DC, Fox PT. Investigating the functional heterogeneity of the default mode network using coordinate-based meta-analytic modeling. J Neurosci. 2009;29:14496–505. 2820256.PubMed CentralPubMedView ArticleGoogle Scholar
- Turner JA, Laird AR. The cognitive paradigm ontology: design and application. Neuroinformatics. 2012;10:57–66. 3682219.PubMed CentralPubMedView ArticleGoogle Scholar
- Erhardt E, Allen E, Damaraju E, Calhoun VD. On network derivation, classification, and visualization: a response to Habeck and Moeller. Brain Connectivity. 2011;1:1–19. PMC Pending #304235.PubMed CentralPubMedView ArticleGoogle Scholar
- Allen E, Erhardt E, Damaraju E, Gruner W, Segall J, Silva R, et al. A baseline for the multivariate comparison of resting state networks. Front Syst Neurosci. 2011;5:12. PMC Journal - In Process.Google Scholar
- Jafri M, Pearlson GD, Stevens M, Calhoun VD. A Method for Functional Network Connectivity Among Spatially Independent Resting-State Components in Schizophrenia. Neuroimage. 2008;39:1666–81. PMC pending #40720.PubMed CentralPubMedView ArticleGoogle Scholar
- Calhoun VD, Adalı T. Multi-subject independent component analysis of fMRI: a decade of intrinsic networks, default mode, and neurodiagnostic discovery. IEEE Reviews in Biomedical Engineering. 2012;5:60–73. PMC23231989.PubMedView ArticleGoogle Scholar
- Damaraju E, Allen EA, Belger A, Ford J, McEwen SC, Mathalon D, et al. Dynamic functional connectivity analysis reveals transient states of dysconnectivity in schizophrenia. Neuroimage Clin. 2014;5:298–308. PMC Journal - In Process.PubMed CentralPubMedView ArticleGoogle Scholar
- SM Plis, D Hjelm, RR Salakhutdinov, EA Allen, HJ Bockholt, J Long, et al. “Deep learning for neuroimaging: a validation study,” Frontiers in Brain Imaging Methods, in press, PMC Journal - In Process.
- D Hjelm, VD Calhoun, RR Salakhutdinov, E Allen, T Adalı, and SM Plis, “Restricted Bolzmann machines for neuroimaging: an application in identifying intrinsic networks,” NeuroImage, in press, PMC Journal - In Process.
- Calhoun VD, Potluru V, Phlypo R, Silva R, Pearlmutter B, Caprihan A, et al. Independent component analysis for brain fMRI does indeed select for maximal independence. PLoS One. 2013;8, MC3757003.View ArticleGoogle Scholar
- Van Dijk KR, Hedden T, Venkataraman A, Evans KC, Lazar SW, Buckner RL. Intrinsic functional connectivity as a tool for human connectomics: theory, properties, and optimization. J Neurophysiol. 2010;103:297–321. 2807224.PubMed CentralPubMedView ArticleGoogle Scholar
- Craddock RC, Jbabdi S, Yan CG, Vogelstein JT, Castellanos FX, Di Martino A, et al. Imaging human connectomes at the macroscale. Nat Methods. 2013;10:524–39.PubMed CentralPubMedView ArticleGoogle Scholar
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.