Abstract
Resting-state functional brain network is defined as temporal correlations between BOLD signals of many brain areas measured by fMRI. Disorder specific abnormal patterns were found in this rs-fcMRI (resting state functional connectivity magnetic resonance imaging) , thus biomarkers of several disorders have been developed based on rs-fcMRI. However, previous biomarkers did not generalize well beyond a single imaging site. We developed a new machine learning technique by combining L1-regularized canonical correlation analysis and sparse logistic regression so that the biomarker achieved more than 80% correct rate for multiple Japanese sites, and the biomarker developed only using Japanese data generalized for US data with more than 70% correct rate (Yahata et al. 2013) . fMRI real-time neurofeedback had been applied as a new therapy for chronic central pain, depression and compulsive disorders. Shibata et al. (2011) developed the “decoded neurofeedback” (DecNef) method by which specific brain activity patterns can be induced noninvasively utilizing fMRI real-time neurofeedback. It is possible to combine the above mentioned biomarker and DecNef to construct a new therapy for psychiatric disorders (Hashimoto et al. 2013) . We ascertained safety of this method and obtained preliminary but encouraging results showing recovery of normal brain network patterns in high-functioning adult autism spectrum disorder patients.
