Multivariate pattern interpretation using PRoNTo

Recently, machine learning models have been applied to neuroimaging data, allowing to make predictions about a variable of interest based on the pattern of activation or anatomy over a set of voxels. In addition, they might lead to an increased sensitivity to detect the presence of a particular mental representation compared to univariate methods such as the General Linear Model (GLM). Application of these methods made it possible to decode the category of a seen object or the orientation of a striped pattern seen by the subject. They also allowed classification of patients and healthy controls and could therefore be used as a diagnostic tool due to their ability to predict the class of an unseen sample.
The main disadvantage of multivariate machine learning models is that local inference with respect to the brain neuroanatomy is complex: although linear models generate weights for each voxel, the model predictions are based on the whole pattern and therefore one cannot threshold the weights to make regional statistical inferences as in univariate analysis. In the present work, we developed a methodology based on a labelled anatomical template (e.g. AAL or Brodmann) to display a smoothed version of the model weights and compute a ranking of the regions according their contribution to the predictive model. This work is distributed in PRoNTo (Pattern Recognition for Neuroimaging Toolbox), a user-friendly toolbox, making machine learning models available to every neuroscientist.