Abstract Summary/Description
Background: Functional neuroimaging has been utilized to study blood flow in the brain. Techniques such as fMRI have been used to characterize these networks in disorders ranging from depression to schizophrenia. Like fMRI, single photon emission computed tomography (SPECT) also captures blood flow activity however through radioactive tracers, and has been used to study psychiatric disorders. While studies utilizing SPECT in schizophrenia populations have been conducted, investigation of aberrant brain networks in SPECT studies is lacking. We implement a novel approach which extracts fMRI-guided networks from SPECT data. We then use subject specific expressions of each network as input to a classifier model to evaluate degree of diagnostic differences in SPECT data. Methods: 213 subjects (137 schizophrenia patients and 76 healthy controls) were used for the analysis. Classification input data was based on loading parameters generated from spatially constrained independent component analysis (ICA) using a set of networks/components derived from fMRI. Fifty-three SPECT components were estimated guided by the NeuroMark fMRI 1.0 template. We then used a support vector machine (SVM) approach which has previously been identified as a useful algorithm for identifying case/control differences in schizophrenia. Results and Conclusion: Linear SVM test results resulted in classifier scores of 83%. Components associated with the auditory, subcortical and sensorimotor networks ranked highest, indicating loading parameters were expressed in brain regions associated with these networks. In conclusion, SPECT data appears to be comparable to resting fMRI data regarding ability to predict individual subject diagnoses.
