Abstract
The sliding-window-based dynamic functional connectivity networks (D-FCNs) derived from resting-state functional magnetic resonance imaging (rs-fMRI) are effective methods for diagnosing various neurological diseases, including autism spectrum disorder (ASD). However, traditional D-FCNs are low-order networks based on pairwise correlation between brain regions, thus overlooking high-level interactions across multiple regions of interest (ROIs). Moreover, D-FCNs suffer from the temporal mismatching issue, i.e., subnetworks in the same temporal window do not have temporal correspondence across different subjects. To address the above problems, we first construct a novel high-order D-FCNs based on the principle of "correlation's correlation" to further explore the higher level and more complex interaction relationships among multiple ROIs. Furthermore, we propose to use a central-moment method to extract temporal-invariance properties contained in either low- or high-order D-FCNs. Finally, we design and train an ensemble classifier by fusing the features extracted from conventional FCN, low-order D-FCNs, and high-order D-FCNs for the diagnosis of ASD and normal control subjects. Our method achieved the best ASD classification accuracy (83%), and our results revealed the features extracted from different networks fingerprinting the autistic brain at different connectional levels.
Introduction
Resting-state functional MRI (rs-fMRI) provides a window into the brain's spontaneous neural activity. By mapping how different brain regions (Regions of Interest, or ROIs) synchronize, researchers create Functional Connectivity Networks (FCNs). These networks are vital for understanding neurological conditions like Autism Spectrum Disorder (ASD).
However, traditional FCN analysis faces two significant challenges. First, most models are "low-order," focusing only on pairwise correlations and missing complex interactions across multiple circuits. Second, "dynamic" snapshots of brain activity often suffer from temporal mismatch; because resting-state scans are unconstrained, a specific time point in one subject does not necessarily correspond to the same state in another. This study introduces a multi-level approach to overcome these hurdles.
Methods
The researchers developed a framework that captures the "fingerprint" of the autistic brain through three distinct levels of network analysis.
Multi-Level Network Construction
Beyond static connectivity, the team employed a sliding-window approach to generate Low-Order Dynamic FCNs (Lo-D-FCNs). To capture even more complex relationships, they calculated a "correlation of correlations," resulting in High-Order Dynamic FCNs (Ho-D-FCNs). This high-order perspective reveals how the connectivity profiles between regions evolve in tandem.
Central-Moment Feature Extraction
To address the temporal mismatch problem, the authors moved away from direct time-series alignment. Instead, they treated connectivity fluctuations as a distribution and extracted central moments (up to the 7th order), such as mean, variance, skewness, and kurtosis. These moments are "order-invariant," meaning they capture the statistical nature of brain fluctuations regardless of when they occur during the scan.
Dataset and Classification
The study used a subset of the ABIDE database from NYU Langone Medical Center, including 45 ASD patients and 47 normal controls (ages 7–15). An ensemble classifier was trained using three separate Linear Support Vector Machines (SVMs) for static, low-order dynamic, and high-order dynamic features, with the final diagnosis being a weighted fusion of these experts.
Results and Discussion
The ensemble method achieved a state-of-the-art classification accuracy of 83%, significantly outperforming conventional static FCNs (74%). Key metrics included a sensitivity of 82% and a specificity of 84%.
The results underscore the power of "correlation's correlation." Ho-D-FCNs identified discriminative biomarkers invisible to standard analysis, particularly in regions associated with emotional expression and motion coordination. By fusing static and dynamic features, the model provides a holistic view of the autistic brain's connectivity.
While successful, future work will focus on more sophisticated fusion strategies and testing the model on larger, multi-site datasets to ensure the generalizability of these central-moment signatures across diverse populations and scanning environments.
Summary updated