COINSTAC: A Decentralized, Differentially Private Application for Neuroimaging

Data sharing is beneficial for science in that it enables researchers to find smaller effects and create more generalized machine learning models. However, data sharing is in some cases difficult or outright impossible because of barriers such as privacy laws, negotiating data usage agreements, infrastructure limitations, or lack of trust. The Collaborative Informatics and Neuroimaging Suite Toolkit for Anonymous Computation (COINSTAC) solves these problems by offering a platform for decentralized and differentially private analyses for neuroimaging researchers. Developed since 2015 by the Joint GSU, Georgia Tech, and Emory Center for Translational Research in Neuroimaging and Data Science (TReNDS) in Atlanta, GA, COINSTAC is a free and open source application in which algorithm developers have implemented various forms of linear regression, classification, and visualization to suit a variety of neuroimaging modalities. COINSTAC enables decentralized regression as well as advanced multivariate and machine learning tools and also enhanced privacy preservation tools such as differential privacy. COINSTAC is an Electron application written in JavaScript that utilizes Docker containers to encapsulate complex neuroimaging and machine learning pipelines. This talk will provide an overview of COINSTAC and show a demonstration of a COINSTAC computation.