Schedule of Events | Symposia

Sketchpad Series

Towards Common Dynamic Connectivity Modes in Schizophrenia and Healthy Controls

Poster Session D - Monday, March 31, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

David Blair¹ (dblair@gsu.edu), Krishna Pusuluri¹, Vince Calhoun¹; ¹Georgia State University

Data analysis has become more time- and resource-intensive than data collection in neuroimaging research. Public datasets allow labs to specialize in analysis, but the lack of methods that address time and spatial dependencies is a major bottleneck. Disposing of temporal information sacrifices detection power for mathematical convenience. We applied independent component analysis constrained by the NeuroMark template [1] (cICA) to the FBIRN Phase III dataset [2], [3], [4] to identify functionally distinct spatial networks. We then computed windowed functional network connectivity (wFNC) and searched for recurring, time-dependent patterns between these networks using dynamic mode decomposition (DMD) [5], [6], [7], [8]. DMD estimates the eigendecomposition of the linear map between consecutive samples, thus accounting for temporal dependencies and providing insights into spatial pattern dynamics, oscillation frequency, and power. All six examined modes showed significant reductions in mean power and variance in schizophrenia. The network-based statistic [9], [10] reveals substantial group-level connectivity changes in all but one mode, with changes concentrating between functional domains. NeuroMark gICA and DMD effectively isolate inter-domain alterations in spatial and temporal structure between groups. The reduced power in the six most dominant modes in patients suggests a flattening of the power distribution or an overall power reduction. Connectivity alterations imply issues with inter-domain integration, supporting the dysconnectivity hypothesis [11]. Both groups display substantial intra-group heterogeneity, indicating possible transdiagnostic subgroups. Future studies should search for such subgroups, perhaps via generative embedding [12] or functional connectivity dynamics-specific tests [13], [14].

Topic Area: METHODS: Neuroimaging