Functional connectivity estimates are highly sensitive to analysis choices and can be dominated by noise when the number of sampled time points is small relative to network dimensionality. This issue is particularly acute in fMRI, where scan resolution is limited. Because scan duration is constrained by practical factors (e.g., motion and fatigue), many datasets remain statistically underpowered for high-dimensional correlation estimation. We introduce a framework that combines diffusion-based structural coarse-graining with spectral noise filtering to recover statistically reliable functional networks from temporally limited data. The method reduces network dimensionality by grouping regions according to diffusion-defined communication. This produces coarse-grained networks with dimensions compatible with available time points, enabling random matrix filtering of noise-dominated modes. We benchmark three common FC pipelines against our approach. We find that raw-signal correlations are strongly influenced by non-stationary fluctuations that can reduce apparent inter-subject variability under limited sampling conditions. In contrast, our pipeline reveals a broader, multimodal landscape of inter-subject variability. These large-scale organization patterns are largely obscured by standard pipelines. Together, these results provide a practical route to reliable functional networks under realistic sampling constraints. This strategy helps separate noise-driven artifacts from reproducible patterns of human brain variability.
Functional connectivity estimates are highly sensitive to analysis choices and can be dominated by noise when the number of sampled time points is small relative to network dimensionality. This issue is particularly acute in fMRI, where scan resolution is limited. Because scan duration is constrained by practical factors (e.g., motion and fatigue), many datasets remain statistically underpowered for high-dimensional correlation estimation. We introduce a framework that combines diffusion-based structural coarse-graining with spectral noise filtering to recover statistically reliable functional networks from temporally limited data. The method reduces network dimensionality by grouping regions according to diffusion-defined communication. This produces coarse-grained networks with dimensions compatible with available time points, enabling random matrix filtering of noise-dominated modes. We benchmark three common FC pipelines against our approach. We find that raw-signal correlations are strongly influenced by non-stationary fluctuations that can reduce apparent inter-subject variability under limited sampling conditions. In contrast, our pipeline reveals a broader, multimodal landscape of inter-subject variability. These large-scale organization patterns are largely obscured by standard pipelines. Together, these results provide a practical route to reliable functional networks under realistic sampling constraints. This strategy helps separate noise-driven artifacts from reproducible patterns of human brain variability.
Complex cognitive functions-such as perception, memory, decision-making, and navigation-emerge from the coordinated activity of large neuronal populations 1,2 . These populations not only encode sensory information but also transmit and integrate it across brain regions to generate appropriate coordinated behavioral responses 3,4 . Remarkably, even in the absence of external stimuli or explicit tasks, the brain dynamics remains spontaneously active [5][6][7] , revealing an interplay in which structural architecture constrains and biases functional interactions without fully determining them [8][9][10] . This intrinsic, non-trivial activity appears to be a fundamental feature of neural computation.
Understanding the origin and functional role of this energy-demanding resting -or “un-resting”-state, and how it interacts with input-driven responses, remains a central challenge in neuroscience 8,11,12 , with implications for how the brain processes and transmits information.
Correlation structure provides a direct window onto functional interactions. By revealing patterns of functional connectivity, correlation analysis facilitated the inference of circuit organization in systems ranging from the retina 13 to the visual thalamocortical pathway 14 and local cortical networks 15,16 . Changes in correlation structure across stimuli or behavioral states can expose computations that single-neuron activity alone would miss [17][18][19][20] .
For example, during active exploration, sensory cortical responses become desynchronized even without changes in input or firing rate 21 . At a larger scale, Functional Magnetic Resonance Imaging (fMRI) enables monitoring of population dynamics across distributed brain regions, providing complementary insight into large-scale functional connectivity. Functional connectivity patterns have also been shown to contain reliable subject-specific signatures across sessions, enabling individual identification from connectome structure [22][23][24] .
In practice, estimating population-level correlations is limited by sampling and noise [25][26][27] . Importantly, a substantial fraction of functional correlations in restingstate fMRI is known to be driven by global, nonneuronal fluctuations related to motion, physiology, and arousal 28,29 . The treatment of these fluctuations remains debated, particularly in the context of global signal regression, which can both reduce artifacts and alter correlation structure 30 . Rather than regressing out global components a priori, our approach targets noise-dominated modes at the spectral level, allowing their statistical identification without imposing a specific preprocessing model.
In high-dimensional datasets, noise can obscure genuine interactions, induce spurious correlations, and cause standard null models to misrepresent the true structure of correlation matrices [31][32][33] . However, fMRI rarely provides arXiv:2602.08910v1 [cond-mat.dis-nn] 9 Feb 2026 the temporal resolution needed for high-dimensional correlation estimation, making it difficult to separate meaningful structure from sampling noise. Reliable covariance estimation typically requires T ≳ N (up to constant factors depending on the estimator and temporal autocorrelation). This limitation is not merely theoretical: empirical studies show that the reliability of functional connectivity estimates depends strongly on scan duration, with shorter acquisitions pr
This content is AI-processed based on open access ArXiv data.