Diagnosis of Autism Spectrum Disorders Using Multi-Level High-Order Functional Networks Derived From Resting-State Functional MRI

Feng Zhao, Han Zhang, Islem Rekik, Zhiyong An, Dinggang ShenFrontiers in Human Neuroscience 2018 · DOI

Abstract

Functional brain networks derived from resting-state functional magnetic resonance imaging (rs-fMRI) have been widely used for Autism Spectrum Disorder (ASD) diagnosis. Typically, these networks are constructed by calculating functional connectivity (FC) between any pair of brain regions of interest (ROIs), i.e., using Pearson's correlation between rs-fMRI time series. However, this can only be called as a low-order representation of the functional interaction, because the relationship is investigated just between two ROIs. Brain disorders might not only affect low-order FC, but also high-order FC, i.e., the higher-level relationship among multiple brain regions, which might be more crucial for diagnosis. To comprehensively characterize such relationship for better diagnosis of ASD, we propose a multi-level, high-order FC network representation that can nicely capture complex interactions among brain regions. Then, we design a feature selection method to identify those discriminative multi-level, high-order FC features for ASD diagnosis. Finally, we design an ensemble classifier with multiple linear SVMs, each trained on a specific level of FC networks, for boosting the final classification accuracy. Experimental results show that the integration of both low-order and first-level high-order FC networks achieves the best ASD diagnostic accuracy (81%). We further investigated those selected discriminative low-order and high-order FC features and found that the high-order FC features can provide complementary information to the low-order FC features in the ASD diagnosis.

Introduction

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by diverse behavioral symptoms and social communication deficits. While resting-state functional MRI (rs-fMRI) has long been used to identify potential biomarkers, traditional methods often rely on Low-Order Functional Connectivity (LON). LON typically measures pairwise interactions between brain regions using Pearson’s correlation. However, ASD may involve more complex, multi-regional "miswiring" that simple pairwise correlations cannot fully capture. This study proposes a novel framework that extracts "high-order" functional relationships—essentially the "correlation of correlations"—to improve diagnostic precision.

Methodology

The proposed framework utilizes a four-step pipeline to transform raw rs-fMRI data into a robust diagnostic tool:

  1. Network Construction: A standard Low-Order Network (LON) is created via Pearson correlation between 116 brain regions. From this, Multi-Level High-Order Networks (HONs) are iteratively derived by correlating the connectivity profiles of different regions rather than their raw time series.
  2. Feature Selection: Given the high dimensionality of these matrices, the Least Absolute Shrinkage and Selection Operator (LASSO) is employed to identify the most discriminative connections.
  3. Ensemble Classification: Separate Linear Support Vector Machines (SVMs) are trained for each level (LON, HON-1, etc.).
  4. Classification Fusion: The final diagnosis is determined by a weighted average of the individual classifier scores.
Overview of the proposed multi-level high-order functional connectivity classification framework for ASD diagnosis.

Results and Discussion

The study evaluated 100 subjects from the ABIDE database (54 ASD, 46 controls). The results demonstrate that high-order networks provide critical, complementary information that traditional methods miss. The ensemble approach combining LON and HON-1 achieved a peak diagnostic accuracy of 81%, significantly outperforming LON alone (73%) or HON-1 alone (70%).

Circular plots and brain maps illustrating the most discriminative features identified across low-order (LON) and high-order (HON) networks.

Biological mapping of the selected features highlighted regions such as the amygdala, hippocampus, and inferior frontal gyrus. These areas are clinically associated with social communication and emotion processing, reinforcing the biological validity of the high-order features. Interestingly, the accuracy decreased at the HON-2 level (67%), suggesting a "sweet spot" for abstraction before the signal becomes too noisy for effective diagnosis.

Conclusion

By moving beyond simple pairwise correlations to multi-level, high-order representations, this framework offers a more comprehensive view of the functional disruptions in ASD. The high-order connectivity profiles serve as robust biomarkers that capture complex topographical interactions. Future research will explore applying this logic to dynamic networks and larger, multi-site datasets to further validate its clinical utility.

Summary updated

Abstract

Functional brain networks derived from resting-state functional magnetic resonance imaging (rs-fMRI) have been widely used for Autism Spectrum Disorder (ASD) diagnosis. Typically, these networks are constructed by calculating functional connectivity (FC) between any pair of brain regions of interest (ROIs), i.e., using Pearson's correlation between rs-fMRI time series. However, this can only be called as a <i>low-order</i> representation of the functional interaction, because the relationship is investigated just between two ROIs. Brain disorders might not only affect low-order FC, but also <i>high-order</i> FC, i.e., the higher-level relationship among multiple brain regions, which might be more crucial for diagnosis. To comprehensively characterize such relationship for better diagnosis of ASD, we propose a multi-level, high-order FC network representation that can nicely capture complex interactions among brain regions. Then, we design a feature selection method to identify those discriminative multi-level, high-order FC features for ASD diagnosis. Finally, we design an ensemble classifier with multiple linear SVMs, each trained on a specific level of FC networks, for boosting the final classification accuracy. Experimental results show that the integration of both low-order and first-level high-order FC networks achieves the best ASD diagnostic accuracy (81%). We further investigated those selected discriminative low-order and high-order FC features and found that the high-order FC features can provide complementary information to the low-order FC features in the ASD diagnosis.
BibTeX
@inproceedings{zhao2018diagnosis,
  author = {Feng Zhao and Han Zhang and Islem Rekik and Zhiyong An and Dinggang Shen},
  title = {{Diagnosis of Autism Spectrum Disorders Using Multi-Level High-Order Functional Networks Derived From Resting-State Functional MRI}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2018},
  doi = {10.3389/fnhum.2018.00184},
}
Figure