Méthodes de prédiction multivariées basées sur la neuroimagerie : application aux maladies psychiatriques

Abstract

The study of mental illness (mood disorders, schizophrenia, autism spectrum disorder) in neuroimaging has been addressed by standard “univariate” statistical analysis methods. These methods have revealed many structural and functional brain changes associated with these diseases. Such analysis answers the question: “Where in the brain and on the scale of a cohort of subjects/patients, are the associations between the pathology and any structural or functional related signal”. This type of analysis has no predictive power to assist in the diagnosis or the prognosis of the response to a treatment at an individual level. In addition, these analyses typically involve examining each voxel separately (measured signal at given locus), thus limiting the identification of “patterns” that jointly involve several brain regions. Taking into account the fact that brain alterations in psychiatric diseases expand over a widely distributed network of brain regions, the neuroimaging community turned to multivariate predictive methods because of their ability to examine voxels jointly and their potential for making inferences at a single-subject level. This article provides a critical overview of the current applications of multivariate prediction methods for the identification of biomarkers based on neuroimaging in the prospect of using these for the diagnosis, early detection and response to treatment of mental illness.