Machine Learning Exploration Of Brain fMRI Data To Study Inhibitory Control Mechanisms

Опубликовано 6 сентября 2016, 16:38

Functional Magnetic Resonance Imaging (fMRI) has enabled scientists to look into the active human brains. This has revealed exciting insights into the spatial and temporal changes underlying a broad range of brain functions, based on a flood of new data thus requiring the development of appropriate data analysis methods. We have recently developed a comprehensive machine learning framework for the exploration of fMRI data, and applied it to a challenging problem: performing classification of hard-to categorize groups of subjects based on simultaneously recorded brain activation response patterns to behavioral challenges of inhibitory control (drug addicted subjects vs. control subjects). The difficulties of the classification problem motivate the joint exploration of spatial, temporal and function information for the analysis of fMRI signals. For spatial analysis, we introduced a novel algorithm that integrates side information into the boosting framework. Our algorithm clearly outperformed well-established classifiers as documented in extensive experimental results. For temporal analysis, we demonstrated that group classification is improved by selecting discriminative features and incorporating fMRI temporal information into a machine learning framework. For functional analysis, we employed Dynamic Bayesian Networks and were able to perform group classification even if the DBNs are constructed from as few as 5 brain regions. Furthermore, different DBN structures characterized drug addicted subjects vs. control subjects. Our results suggest that through incorporation of machine learning principles into functional neuroimaging studies we will be able to identify unique patterns of variability in brain states and deduce about the behavioral probes from the brain activation data. In the future this may provide tools where objective brain imaging data are used for clinical purpose of classification of psychopathologies and identification of genetic vulnerabilities.