Machine learning in neuroimaging: development of methods and clinical applications in psychiatric disorders

Abstract

Neuroimaging methods technology such as functional MRI, tractography, voxel-based morphometry, analysis of cortical surface are advancing rapidly, allowing the observation of distinct aspects of brain structure and function, playing a fundamental contribution to the understanding of mental disorders (MD). Traditionally, "average" comparison between groups of healthy subjects and patients with MD has contributed to the enhance the comprehension of the ethiology and neural substrates of pathologies. Since the changes are subtle and variability of the data is large, there is an overlap between the two groups. This overlap strongly constrains the use of neuroimaging techniques in clinical practice, since they can not be applied to single individual diagnosis. Thus, a major challenge in psychiatry is the discovery of biomarkers such as neuroimaging, which are useful both for diagnosis of TM and to predict the efficacy of treatments with drugs and / or therapies (prognosis). Subtle brain changes are spatially distributed and difficult to detect with the naked eye but can be detected with the use of machine learning methods. In previous work of our group, we could demonstrate that some of these techniques allow to obtain predictions of mental states (e.g., pathological or refractory) to the individual case. In this project, we want to make a conceptual jump in development of methods and applications in psychiatry. This jump underlies on the use of large neuroimaging non-public databases (on the order of hundreds of individuals) to be combined with genetic (DNA and RNA) to validate the techniques previously developed and perform data mining to discover markers for psychiatric disorders. We hope to find potential biomarkers for schizophrenia through measures of connectivity and genetic polymorphisms. Furthermore, we study trajectories of functional connectivity to identify potential early markers of risk for MD in a sample (unprecedented in the world) including more than 750 children with healthy development and children at high risk for TM. Our group has been successful in developing methodologies combining advanced methods of pattern recognition neuroimaging to MD. This project is in accordance of current literature which reinforces the relevance of technological advances for diagnostics and prognosis in psychiatry. (AU)