Fusion of Novel FMRI Features Using Independent Vector Analysis for a Multifaceted Characterization of Schizophrenia

Abstract

The fractional amplitude of low-frequency fluctuation (fALFF) is a widely used feature for resting-state functional magnetic resonance (fMRI) analysis but captures limited information. Here, we propose two novel features, maxTP (maximum amplitude across time points) and max-RSN (maximum values across resting-state networks), that capture temporal peaks and salient spatial components, respectively. Using fMRI data from the Bipolar and Schizophrenia Network for the Intermediate Phenotypes project, we constructed a dataset by combining fALFF with the proposed features. Subsequently, we applied a data fusion framework by utilizing independent vector analysis on this dataset, leveraging both second and higherorder statistical information. Our analysis revealed significant group differences between schizophrenia patients and healthy controls in various brain regions. Notably, differences in the visual cortex were detected across all three feature datasets, suggesting its potential as a schizophrenia biomarker across different measures. Thus, by incorporating new features and a multi-feature data fusion approach, this study provides insights into the multifaceted nature of brain alterations in schizophrenia, emphasizing the importance of conducting neuroimaging analyses with complementary features to explore brain activity changes in psychiatric conditions.