Multi-model Order ICA: A Data-driven Method for Evaluating Brain Functional Network Connectivity Within and Between Multiple Spatial Scales

Xing Meng, Armin Iraji, Zening Fu, Peter Kochunov, Ayşenil Belger, Judy M. Ford, Sara McEwen, Daniel H. Mathalon, Bryon A. Mueller, Godfrey D. Pearlson, Steven G. Potkin, Adrian Preda, Jessica A. Turner, Theo G.m. van Erp, Jing Sui, Vince D. CalhounbioRxiv 2021 · DOI

TL;DR

This paper introduces a novel "multi-model order ICA" approach to analyze brain functional network connectivity (FNC) across different spatial scales. By examining FNC at various levels of detail (model orders), the researchers found unique insights into schizophrenia that would be missed with traditional single-scale analyses, demonstrating its potential for improved brain disorder classification.

What Problem Does This Paper Address?

Functional connectivity (FC) is a key area of neuroscience research, but most studies analyze it at a single spatial scale. This means we might be missing important information about how brain networks interact at different levels of granularity, from small, localized networks to large, distributed ones. The relationship between FC at these different scales and how it might change in neurological conditions like schizophrenia remains largely unknown. This paper aims to bridge this gap by developing a method to explore FNC across multiple spatial scales simultaneously.

Methods

The core of this work is a novel application of Independent Component Analysis (ICA) called "multi-model order ICA." Standard ICA decomposes brain activity into a set of independent components (ICs), which represent functional networks. The "model order" in ICA refers to the number of components extracted. By running ICA at multiple model orders (specifically 25, 50, 75, and 100 in this study), the researchers can capture functional networks at different spatial resolutions. They then analyze the Functional Network Connectivity (FNC) – the correlation between these ICs – both within each model order and between different model orders. To assess the clinical relevance, they used Support Vector Machine (SVM) classification to see how well these multi-scale FNC features could distinguish between individuals with schizophrenia (SZ) and healthy controls (HC).

Datasets

The study utilized resting-state fMRI (rsfMRI) data from a combined dataset of 827 individuals, comprising 477 healthy controls (HC) and 350 individuals with schizophrenia (SZ). This data was pooled from three separate studies: FBIRN (seven sites), MPRC (three sites), and COBRE (one site). The paper does not introduce new benchmarks but rather uses existing, well-characterized datasets for its analysis.

Key Insights and Ideas

  • Multi-scale FNC reveals unique information: The most significant insight is that analyzing FNC across multiple spatial scales (model orders) uncovers unique predictive information that is missed when looking at only one scale. This suggests that brain function is inherently hierarchical and that understanding interactions at different levels is crucial.
  • Interactions between scales matter: The study highlights that not only FNC within a specific scale but also the FNC between different scales (i.e., how networks at one resolution relate to networks at another) holds important diagnostic information.
  • Specific domains are key: The somatomotor and visual domains consistently showed strong predictive power for differentiating between HCs and SZs, both within and across model orders. This points to specific functional systems being particularly affected in schizophrenia.
  • Lower model orders offer unique insights: Interestingly, some group differences (HC vs. SZ) were more pronounced at lower model orders (e.g., subcortical vs. temporal/cerebellum), while others (somatomotor vs. visual) were more apparent at higher model orders. This reinforces the idea that different spatial scales capture different aspects of the disorder.

Key Results

  • Classification Performance: The SVM model achieved an average accuracy of 77.4% in classifying HCs from SZs, outperforming a null accuracy of 57.6%.
  • Predictive Features: FNC features within the somatomotor and visual domains were particularly strong contributors to classification across all model orders.
  • Model Order Importance: Features from between-model orders (specifically 25-50) were found to be consistently more predictive than features within individual model orders. Higher individual model orders generally led to higher classification accuracies.
  • Shared Templates: The study provides shared spatial templates of ICNs at different model orders, which can serve as a valuable resource for future research using ICA-based approaches.

Limitations and Future Work

  • Feature Selection Strategy: The authors acknowledge that the feature selection process, while aiming for stability, involved a strategy similar to bootstrap sampling. Future work could explore increasing the number of feature selection rounds or varying the subset size to potentially achieve a more stable feature set.
  • Machine Learning Exploration: While SVM performed well, the authors experimented with random forests and found SVM slightly better for their specific task. They suggest that exploring other machine learning techniques like boosting could further improve classifier accuracy.
  • Further Investigation of Between-Order FNC: The study emphasizes the importance of between-model order FNC. Future research could delve deeper into understanding the specific mechanisms and interpretations of these cross-scale interactions.

Summary updated

Abstract

Abstract Background While functional connectivity is widely studied, there has been little work studying functional connectivity at different spatial scales. Likewise, the relationship of functional connectivity between spatial scales is unknown. Methods We proposed an independent component analysis (ICA) - based approach to capture information at multiple model orders (component numbers) and to evaluate functional network connectivity (FNC) both within and between model orders. We evaluated the approach by studying group differences in the context of a study of resting fMRI (rsfMRI) data collected from schizophrenia (SZ) individuals and healthy controls (HC). The predictive ability of FNC at multiple spatial scales was assessed using support vector machine (SVM)-based classification. Results In addition to consistent predictive patterns at both multiple-model orders and single model orders, unique predictive information was seen at multiple-model orders and in the interaction between model orders. We observed that the FNC between model order 25 and 50 maintained the highest predictive information between HC and SZ. Results highlighted the predictive ability of the somatomotor and visual domains both within and between model orders compared to other functional domains. Also, subcortical-somatomotor, temporal-somatomotor, and temporal-subcortical FNCs had relatively high weights in predicting SZ. Conclusions In sum, multi-model order ICA provides a more comprehensive way to study FNC, produces meaningful and interesting results which are applicable to future studies. We shared the spatial templates from this work at different model orders to provide a reference for the community, which can be leveraged in regression-based or fully automated (spatially constrained) ICA approaches. Impact Statement Multi-model order ICA provides a comprehensive way to study brain functional network connectivity within and between multiple spatial scales, highlighting findings that would have been ignored in single model order analysis. This work expands upon and adds to the relatively new literature on resting fMRI-based classification and prediction. Results highlighted the differentiating power of specific intrinsic connectivity networks on classifying brain disorders of schizophrenia patients and healthy participants, at different spatial scales. The spatial templates from this work provide a reference for the community, which can be leveraged in regression-based or fully automated ICA approaches.
BibTeX
@misc{meng2021multi,
  author = {Xing Meng and Armin Iraji and Zening Fu and Peter Kochunov and Ayşenil Belger and Judy M. Ford and Sara McEwen and Daniel H. Mathalon and Bryon A. Mueller and Godfrey D. Pearlson and Steven G. Potkin and Adrian Preda and Jessica A. Turner and Theo G.M. van Erp and Jing Sui and Vince D. Calhoun},
  title = {{Multi-model Order ICA: A Data-driven Method for Evaluating Brain Functional Network Connectivity Within and Between Multiple Spatial Scales}},
  year = {2021},
  doi = {10.1101/2021.10.24.465635},
  howpublished = {bioRxiv},
}
Figure