Mapping longitudinal scientific progress, collaboration and impact of the Alzheimer's disease neuroimaging initiative

Abstract

Background: Alzheimer's disease neuroimaging initiative (ADNI) is a landmark imaging and omics study in AD. ADNI research literature has increased substantially over the past decade, which poses challenges for effectively communicating information about the results and impact of ADNI-related studies. In this work, we employed advanced information visualization techniques to perform a comprehensive and systematic mapping of the ADNI scientific growth and impact over a period of 12 years.

Methods: Citation information of ADNI-related publications from 01/01/2003 to 05/12/2015 were downloaded from the Scopus database. Five fields, including authors, years, affiliations, sources (journals), and keywords, were extracted and preprocessed. Statistical analyses were performed on basic publication data as well as journal and citations information. Science mapping workflows were conducted using the Science of Science (Sci2) Tool to generate geospatial, topical, and collaboration visualizations at the micro (individual) to macro (global) levels such as geospatial layouts of institutional collaboration networks, keyword co-occurrence networks, and author collaboration networks evolving over time.

Results: During the studied period, 996 ADNI manuscripts were published across 233 journals and conference proceedings. The number of publications grew linearly from 2008 to 2015, so did the number of involved institutions. ADNI publications received much more citations than typical papers from the same set of journals. Collaborations were visualized at multiple levels, including authors, institutions, and research areas. The evolution of key ADNI research topics was also plotted over the studied period.

Conclusions: Both statistical and visualization results demonstrate the increasing attention of ADNI research, strong citation impact of ADNI publications, the expanding collaboration networks among researchers, institutions and ADNI core areas, and the dynamic evolution of ADNI research topics. The visualizations presented here can help improve daily decision making based on a deep understanding of existing patterns and trends using proven and replicable data analysis and visualization methods. They have great potential to provide new insights and actionable knowledge for helping translational research in AD.

Fig 1. Statistics for ADNI publications between 01/01/2003 to 05/12/2015.

(A) Growth of ADNI publications on the year-by-year basis; line indicates a linear regression prediction for the 2015 number using data from 2008 to 2014. (B) Growth of institutions involved in ADNI publications; line indicates a linear regression prediction for the 2015 number using data from 2008 to 2014. (C) Distribution of number of authors per paper. (D) Distribution of number of institutions per paper.

Fig 2. Plot of citation counts.

The combined bar and line graph compares the expected citation counts (green bars) and total article citation counts (blue bars) though May 2015 for publications grouped by their annual journal SRJ publication groups (see S1 Table for detailed information about each group). Expected citations were calculated by multiplying the corresponding annual journal IPP score by the number of publications, and summing the totals for each SJR group. Total citations are calculated as the sum of citation counts provided by the Scopus database at the date of retrieval. The number of unique journals per group and the minimum and maximum SJR ranks are provided, as are equivalent Impact Factor scores calculated using a predictive equation generated by regression analysis of ADNI venue Impact Factor and SJR values between 2003 and 2015.

Fig 3. Geospatial map of publication co-occurrence network.

Co-affiliation network overlaid on a geospatial map shows collaborating organizations affiliated with ADNI in North American based on co-authored publications. Only organizations with at least 4 publications are shown; organizations with at least 30 publications or that are a Core ADNI research institution have been labeled in the network. Organization relationships (edges) with four or more co-authorships are shown.

Fig 4. Keyword co-occurrence networks.

(A) Keyword co-occurrence network containing only nodes with degree ≥ 10. Nodes represent keywords, and edges denote the joint appearance of keywords in a publication. Only nodes with degree ≥ 10 are shown. Both nodes and edges are scaled proportionally based on Bezier curve. Nodes are colored based on their categories: genotype, phenotype, analysis, and others. (B) Temporal profiles of selected keywords show the normalized frequency (frequency of keywords divided by number of publications) from 2008 to 2014. Hierarchical plot indicates clustering of selected keywords.

Fig 5. Growth of co-publication networks over time.

Nodes represent institutions. Nodes are colored based on modularity and sized proportional to the natural logarithm of their degrees. Edges represent co-occurrence of institutions in publications, and are sized proportional to the number of co-publications. See S8 Fig for detailed sub networks containing only hub institutions with degree ≥ 30.