Improving Sparsity and Modularity of High-Order Functional Connectivity Networks for MCI and ASD Identification

Yueying Zhou, Limei Zhang, Shenghua Teng, Lishan Qiao, Dinggang ShenFrontiers in Neuroscience 2018 · DOI

Abstract

High-order correlation has recently been proposed to model brain functional connectivity network (FCN) for identifying neurological disorders, such as mild cognitive impairment (MCI) and autism spectrum disorder (ASD). In practice, the high-order FCN (HoFCN) can be derived from multiple low-order FCNs that are estimated separately in a series of sliding windows, and thus it in fact provides a way of integrating dynamic information encoded in a sequence of low-order FCNs. However, the estimation of low-order FCN may be unreliable due to the fact that the use of limited volumes/samples in a sliding window can significantly reduce the statistical power, which in turn affects the reliability of the resulted HoFCN. To address this issue, we propose to enhance HoFCN based on a regularized learning framework. More specifically, we first calculate an initial HoFCN using a recently developed method based on maximum likelihood estimation. Then, we learn an optimal neighborhood network of the initially estimated HoFCN with sparsity and modularity priors as regularizers. Finally, based on the improved HoFCNs, we conduct experiments to identify MCI and ASD patients from their corresponding normal controls. Experimental results show that the proposed methods outperform the baseline methods, and the improved HoFCNs with modularity prior consistently achieve the best performance.

Introduction

Identifying neurological disorders like Mild Cognitive Impairment (MCI) and Autism Spectrum Disorder (ASD) relies heavily on mapping brain functional connectivity through resting-state fMRI (rs-fMRI). Traditional "low-order" models measure simple correlations between brain regions. However, "high-order" models (HoFCNs) have emerged to capture the dynamic "correlation of correlations" over time. While promising, these models often suffer from noise introduced by the "sliding window" approach used to capture temporal dynamics. This paper introduces a novel framework to improve the reliability of HoFCNs by applying sparsity and modularity priors, mimicking the brain's natural biological organization.

Methods

The proposed framework enhances HoFCNs through a two-step regularized learning process:

  1. Initial Estimation: An initial HoFCN is calculated using a Maximum Likelihood Estimation (MLE) approach. This provides a baseline that captures high-order interactions but remains susceptible to noise from short-duration sliding windows.
  2. Neighborhood Network Learning: The initial network is optimized by solving for a new network that remains close to the MLE estimate while adhering to two key mathematical constraints:
  3. Sparsity ($l_1$-norm): This regularizer forces weak, likely spurious connections to zero, resulting in the S-HoFCN model.
  4. Modularity (Trace norm): This encourages the formation of dense functional clusters (modules) with sparse inter-module connections, resulting in the M-HoFCN model.
Comparison of adjacency matrices for different FCN construction methods.

Experimental Evaluation

The methodology was evaluated on two major benchmarks: the ADNI dataset (137 subjects for MCI identification) and the ABIDE dataset (92 subjects for ASD identification). Brain regions were defined using the Automated Anatomical Labeling (AAL) atlas with 116 regions of interest (ROIs).

Key Results

The M-HoFCN model, which incorporates both sparsity and modularity, consistently outperformed baseline methods:

  • MCI Identification: M-HoFCN achieved 86.13% accuracy, significantly higher than traditional Pearson’s Correlation (79.56%) and the un-optimized HoFCN-MLE (82.48%).
  • ASD Identification: M-HoFCN reached 75.00% accuracy, compared to only 63.04% for traditional low-order methods.
  • Robustness: The modularity-regularized networks showed greater stability across various sliding window sizes, reducing the sensitivity to parameter selection.

Discussion

The study reveals that modularity is a more powerful prior than simple sparsity for brain network modeling. While sparsity effectively removes noise, modularity captures the underlying structural "truth" of neural organization. High-order networks appear to reveal finer-grained modular structures that are often obscured in traditional low-order models.

However, challenges remain. The high dimensionality of HoFCNs (over 6,000 features for 116 ROIs) poses a risk of overfitting in small sample sizes. Furthermore, performance was found to be sensitive to the choice of brain atlas; increasing the number of ROIs to 200 actually decreased diagnostic accuracy, likely due to an unfavorable feature-to-sample ratio.

Conclusion

By integrating biological priors into the estimation of high-order functional connectivity, this research provides a more robust and accurate tool for diagnosing neurological disorders. Future work will focus on multi-layer graph clustering and more sophisticated ROI definitions to further refine these diagnostic models.

Summary updated

Abstract

High-order correlation has recently been proposed to model brain functional connectivity network (FCN) for identifying neurological disorders, such as mild cognitive impairment (MCI) and autism spectrum disorder (ASD). In practice, the high-order FCN (HoFCN) can be derived from multiple low-order FCNs that are estimated separately in a series of sliding windows, and thus it in fact provides a way of integrating dynamic information encoded in a sequence of low-order FCNs. However, the estimation of low-order FCN may be unreliable due to the fact that the use of limited volumes/samples in a sliding window can significantly reduce the statistical power, which in turn affects the reliability of the resulted HoFCN. To address this issue, we propose to enhance HoFCN based on a regularized learning framework. More specifically, we first calculate an initial HoFCN using a recently developed method based on maximum likelihood estimation. Then, we learn an optimal neighborhood network of the initially estimated HoFCN with sparsity and modularity priors as regularizers. Finally, based on the improved HoFCNs, we conduct experiments to identify MCI and ASD patients from their corresponding normal controls. Experimental results show that the proposed methods outperform the baseline methods, and the improved HoFCNs with modularity prior consistently achieve the best performance.
BibTeX
@inproceedings{zhou2018improving,
  author = {Yueying Zhou and Limei Zhang and Shenghua Teng and Lishan Qiao and Dinggang Shen},
  title = {{Improving Sparsity and Modularity of High-Order Functional Connectivity Networks for MCI and ASD Identification}},
  booktitle = {Frontiers in Neuroscience},
  year = {2018},
  doi = {10.3389/fnins.2018.00959},
}
Figure