Multivariate pattern classification on BOLD activation pattern induced by deep brain stimulation in motor, associative, and limbic brain networks

Abstract

ABSTRACTFunctional magnetic resonance imaging (fMRI) concurrently conducted with the deep brain stimulation (DBS) has shown that diffuse BOLD activation occurred not only near stimulation locus, but in multiple brain networks, supporting that network-wide modulation would underlie its therapeutic effect. While the extent and pattern of activation varies depending on specific anatomical locus stimulated by DBS, some stimulation targets could induce similar activation pattern in cerebral cortex, albeit different therapeutic and adverse effects were yielded.In order to characterize the unique network-level activation effects of three DBS targets (subthalamic nucleus, the globus pallidus internus, and the nucleus accumbens), we trained the pattern classifier with DBS-fMRI data from three stimulation groups (21 healthy swine), wherein five six seconds of electrical stimulation was conducted while gradient-echo echo planar imaging was on going. Then whole brain regions were systematically grouped into different size of network-of-interest and the classification accuracy for individual target region was quantitatively assessed. We demonstrated that the pattern classifier could successfully differentiate BOLD activation pattern of cortical and subcortical brain regions originated from each individual stimulation target. Moreover, the success rate of classification indicated that some brain regions evoked indistinguishable BOLD pattern, suggesting the presence of commonly activated regions, which was influenced by stimulating different DBS targets.Our results provide an understanding of the biomarker of BOLD pattern that is associated with clinical effectiveness as well as an adverse effect associated to the stimulation. Further, we provide the proof-of-concept for multivariate pattern analysis that is capable of disentangling the complicated BOLD activation pattern, which cannot be readily achieved by a conventional univariate analysis.