Toward Meta-connectomic Ergodicity in Neuroimaging: Theoretical and Experimental Motivation for Developing Data-driven Biomarkers
Jonathan Towne1, Vahid Eslami3, Heath Pardoe4, Jose Cavazos2, Peter Fox1
1Research Imaging Institute, UT Health San Antonio, 2UT Health San Antonio, 3UCLA Health, 4The Florey Institute of Neuroscience and Mental Health
Objective:
To provide evidence of ergodicity in the patho-connectome of temporal lobe epilepsy and to motivate meta-connectomic biomarker development for translational/clinical applications.
Background:
Ergodicity in a dynamical system asserts that group observations at a single time point are equivalent to a single-individual observation over time. In the brain, this would mandate that network properties derived from cross-sectional data will be observed longitudinally in individuals. The implication for neuroimaging, if ergodicity holds, is that meta-analytic sampling can access network architecture (data structures) useful for detecting networks per-subject.
Ergodicity has been implicitly shown in healthy subjects by graph theory (Crossley et al., 2013) and other analytics (Smith et al., 2009). Equivalent functional architecture was identified by connectomic meta-analysis of task-based studies and connectomic analysis of temporally concatenated rs-fMRI. Diseases follow collectively similar yet individually distinct patterns (Vanasse et al., 2021), motivating ergodic hypotheses. We present evidence of ergodicity in a temporal lobe epilepsy (TLE) cohort.
Design/Methods:
Meta-connectomic and connectomic models were derived from published coordinate data from case-control contrasts (n=74 TLE experiments) and from primary (per-subject) rs-fMRI (n=37 TLE/19 controls), respectively. Models were compared by modularity analysis and node topology metrics (e.g. centrality).
Results:
TLE networks identified cross-sectionally (case-control contrasts) were observed longitudinally in TLE, not controls. Two TLE modules were found meta-analytically (limbic & language networks) and present individually. The medial dorsal nucleus was the most topologically influential node; other common hubs included the hippocampus, caudate body, superior temporal gyrus, & inferior parietal lobule.
Conclusions:
Ergodicity was demonstrated in TLE. Critics purport ergodicity to imply individuals are identical. We suggest network structure is similar cross-sectionally (mean coherent structure) but exhibited ergodically over time. These results motivate the application of meta-analytic functional network models in primary data, to develop per-subject biomarkers.