Decoding Brain Dynamics Via Cross-Network Non-Linear Trajectories

Masoud Seraji, Sarah Shultz, Qiang Li, Liang Ma, Zening Fu, Vince D. CalhounbioRxiv 2025 · DOI

TL;DR

This paper introduces a novel method, Cross-Network Non-linear Trajectory Characterization (CNTC), to compare how different brain networks mature during infancy. By fitting polynomial models to network properties over time, they reveal that visual and cerebellar systems develop much faster than motor and attention networks, offering a new way to benchmark typical and potentially atypical brain development.

What Problem Does This Paper Address?

Understanding how the brain develops is crucial, especially during infancy when fundamental functional networks are rapidly emerging and reorganizing. While previous research has mapped brain networks using resting-state fMRI (rsfMRI) and shown that different networks mature at different rates, a significant gap exists in directly comparing these developmental trajectories across networks. Most studies focus on individual networks or use simple summaries, making it difficult to understand the relative timing and shape of maturation for various brain systems. This lack of cross-network comparison hinders our ability to identify early biomarkers for atypical development.

Methods

The core innovation here is the Cross-Network Non-linear Trajectory Characterization (CNTC). This method involves:

  1. Network Estimation: Using group Independent Component Analysis (ICA) on rsfMRI data to identify canonical brain networks.
  2. Spatial Metrics: For each identified network and subject, three spatial measures are calculated:
  3. Network-Averaged Spatial Similarity (NASS): How closely an individual network's spatial map matches the group's average template.
  4. Network Strength: The average contribution of voxels exceeding a certain threshold (indicating engagement).
  5. Network Size: The spatial extent (number of voxels) of the network above the threshold.
  6. Polynomial Modeling: For each of the three spatial metrics within each network, a cubic polynomial is fitted as a function of corrected age (in days). This model takes the form: $f(t) = a_0 + a_1t + a_2t^2 + a_3t^3$.
  7. The coefficients ($a_1, a_2, a_3$) capture the linear (steady change), quadratic (acceleration/deceleration), and cubic (S-shaped dynamics) components of the developmental trajectory.
  8. Cross-Network Contrasts: The key step is comparing these polynomial coefficients between different networks. This allows researchers to quantify differences in the shape and rate of maturation across systems. Statistical significance is assessed using z-statistics and corrected for multiple comparisons (FDR q < 0.05).

This approach moves beyond simply observing trends to quantitatively comparing the dynamics of maturation across distinct brain systems.

Datasets

The study utilized a longitudinal dataset of resting-state fMRI (rsfMRI) from 74 neurotypically developing infants. These infants were scanned between 0 and 6 months of age, with a total of 137 scans. The dataset included infants with no familial autism, developmental delay, perinatal complications, or other neurological/genetic disorders. Age was corrected to a 40-week gestational baseline.

Key Insights and Ideas

  • Differential Maturation Rates: The most striking insight is the clear divergence in maturation speeds between different brain systems. Visual and cerebellar networks show a much steeper and faster developmental trajectory compared to motor and attention networks.
  • Polynomials Capture Complexity: The use of cubic polynomials effectively captures the non-linear nature of brain development, revealing that simple linear models would be insufficient. Quadratic terms (curvature) were particularly important, suggesting accelerations and decelerations in growth.
  • CNTC as a Benchmark: The CNTC framework provides a compact, coefficient-level summary of inter-network maturation. This offers a standardized way to compare developmental trajectories, which can be invaluable for identifying deviations in atypical development.
  • Size vs. Strength vs. Similarity: While all metrics showed age-related changes, network size exhibited the richest cross-network divergence, suggesting that the physical expansion of networks follows distinct patterns.

Key Results

  • Robust Age Effects: All seven canonical systems analyzed showed significant age-related changes in NASS, network strength, and network size.
  • Dominance of Linear and Quadratic Terms: Cross-network differences were primarily driven by linear (slope) terms, indicating differences in the rate of convergence to adult-like patterns. Quadratic terms were also significant, especially for network size, highlighting differences in the curvature of growth. Cubic terms were rare, suggesting that full S-shaped dynamics are less common in this early developmental window.
  • Visual and Cerebellar Acceleration: The primary visual and cerebellar systems showed the steepest NASS slopes and pronounced increases in strength and size, indicating rapid early maturation.
  • Gradual Motor/Attention Development: Motor and attention networks exhibited more gradual changes, with slower increases in NASS and less dramatic growth in strength and size.
  • Network Size Divergence: Network size showed the most significant cross-network differences, with substantial expansion in visual and cerebellar systems compared to more modest changes in other networks.

Limitations and Future Work

  • Limited Age Range: The study focused on a relatively narrow age window (0-6 months). The authors suggest that future work could extend this analysis to cover a broader developmental span to capture later maturation stages.
  • Atypical Development: While CNTC is proposed as a basis for benchmarking atypical development, this study focused solely on neurotypical infants. Future research should apply CNTC to clinical populations to validate its utility in identifying developmental disorders.
  • Metric Selection: The study used three specific spatial metrics. Exploring other network measures or functional connectivity metrics could provide a more comprehensive understanding of developmental dynamics.
  • Cubic Term Rarity: The limited significance of cubic terms might suggest that the 0-6 month period is dominated by linear and quadratic growth, but further investigation across a wider age range is warranted.

Summary updated

Abstract

ABSTRACT Infant brain networks mature rapidly and nonlinearly, yet cross-network comparisons of developmental trajectories remain rare. We introduce cross-network non-linear trajectory characterization (CNTC), which fits cubic polynomials to age-related change in three spatial measures—network-averaged spatial similarity (NASS), network strength, and network size—derived from resting-state fMRI. Using 137 scans from 74 neurotypical infants spanning 0–6 months of age, we estimated networks via group ICA, modeled age trajectories within each network, and then contrasted coefficients across networks. Across seven canonical systems, trajectories showed robust age effects. Cross-network differences were dominated by linear terms, with selective quadratic effects and few cubic effects. Visual and cerebellar systems exhibited the steepest NASS slopes and pronounced increases in strength and size, whereas motor/attention changes were more gradual. CNTC provides a compact, coefficient-level summary of inter-network maturation and a basis for benchmarking atypical development in future work.
BibTeX
@misc{seraji2025decoding,
  author = {Masoud Seraji and Sarah Shultz and Qiang Li and Liang Ma and Zening Fu and Vince D. Calhoun},
  title = {{Decoding Brain Dynamics Via Cross-Network Non-Linear Trajectories}},
  year = {2025},
  doi = {10.1101/2025.11.25.690593},
  howpublished = {bioRxiv},
}
Figure