Can we interpret linear kernel machine learning models using anatomically labelled regions
Jessica Schrouff, João M. Monteiro, Maria João Rosa, Liana Catarina Lima Portugal, Christophe L. M. Phillips, Janaı́na Mourão-Miranda · ORBi (University of Liège) · 2014
Conclusions:While machine learning models allow the prediction of a variable of interest, localizing the information leading to the prediction is complex due to their multivariate nature.In this work, we propose to use a priori anatomical information to build sparse hierarchical multivariate models and thereby facilitate model interpretation.Although the proposed approach depends on the precision and resolution of the anatomical template, the framework is general and can be applied to different templates.The methods were implemented in PRoNTo [9], which is a Matlab-based, SPM compatible toolbox.