Ref: automatic meta-analysis (http://neurosynth.org/)

Ref: http://www.nature.com/nmeth/journal/v8/n8/full/nmeth.1661.html?WT.ec_id=NMETH-201108

http://neurosynth.org

From journal articles to computational models: a new automated tool

• Tom M. Mitchell1

Journal name:

Nature Methods

Volume:

8,

Pages:

627–628

Year published:

(2011)

DOI:

doi:10.1038/nmeth.1661

Published online

28 July 2011

Automated methods can now extract brain-image coordinates appearing in hundreds of publications in targeted topic areas and then integrate these data to form computational models that classify new brain-image data.

 

Most neuroscience research today is data-starved: it seeks to model a very complicated system (the brain) using only the data from a single experimental study. As we attempt to build increasingly accurate models of larger and larger brain systems, the only plausible way forward is to build models that integrate data and distilled results from many previously published studies, along with the new experimental data from our own work. One approach will be to build shared repositories of curated experimental data, and such efforts have already begun. Today, however, the largest already available shared repository of experimental results is in online journals.

In a paper in this issue, Yarkoni et al.1 present an important new approach to integrate functional magnetic resonance imaging (fMRI) experimental results published in thousands of online journal articles. They describe a method and accompanying web service (http://neurosynth.org/) that enables neuroscientists to search for online articles by various psychological keywords (for example, pain) and that automatically extracts brain coordinates from tables in these articles. The program then uses machine-learning techniques to perform automatic meta-analysis of these extracted data (Fig. 1). Whereas earlier approaches to meta-analysis tend to be labor-intensive, this automated method enables scientists to quickly perform different meta-analyses based on different search queries over the collection of journal articles.

Figure 1: Automated machine-learning–based meta-analysis of fMRI data.

A computer scans thousands of journal articles for a specific keyword or phrase a researcher is interested in (such as 'working memory') and computes a map of the brain showing the probability of any region being associated with it.

......