Estimated connectivity networks outperform observed connectivity networks when classifying people with multiple sclerosis into disability groups

Ceren Tozlu, Keith Jamison, Zijin Gu, Susan A. Gauthier, Amy KuceyeskibioRxiv 2021 · DOI

TL;DR

This study demonstrates that connectivity networks estimated solely from lesion masks in multiple sclerosis (MS) patients can classify disability levels more effectively than networks derived from expensive, advanced MRI scans. This "estimated" approach offers a more accessible and potentially more accurate way to understand how brain damage in MS relates to functional deficits.

What Problem Does This Paper Address?

Multiple Sclerosis (MS) is a debilitating disease that disrupts the brain's structural and functional networks, leading to various disabilities. Understanding how these disruptions cause disability is crucial for diagnosis, prognosis, and treatment. While advanced MRI techniques like diffusion and functional MRI can measure these brain networks (structural connectivity - SC, and functional connectivity - FC), they are costly, time-consuming, and require specialized expertise, making them impractical for widespread clinical use. This paper addresses the challenge of finding a more accessible method to map brain damage to disability in MS patients. They propose using the Network Modification (NeMo) Tool, which estimates SC and FC (eSC and eFC) using only conventional MRI-derived lesion masks and a reference dataset of healthy controls. The core question is whether these "estimated" connectivity networks can perform as well as, or better than, directly measured SC and FC in classifying MS patients into different disability groups.

Methods

The researchers employed a multi-faceted approach:

  • Observed Connectivity: They extracted SC and FC directly from advanced MRI (diffusion and resting-state fMRI) of MS patients.
  • Estimated Connectivity (NeMo Tool):
    • eSC: The NeMo Tool uses a patient's lesion mask to remove streamlines from a reference tractography dataset (from the Human Connectome Project), effectively estimating structural disconnectivity caused by lesions.
  • eFC: They extended the NeMo Tool by developing a deep neural network trained to predict FC from SC. This network takes the eSC as input and outputs an estimated FC (eFC).
  • Classification: They used logistic regression with ridge regularization to classify patients into "no disability" vs. "evidence of disability" groups. This classification was based on demographic/clinical information and either pairwise connectivity or regional node strength from SC, FC, eSC, and eFC matrices.
  • Performance Evaluation: The Area Under the ROC Curve (AUC) was the primary metric for assessing classification performance. They also used permutation tests to compare model performances and identify significant differences.
  • Feature Importance: Univariate statistics and model coefficients were used to identify the most important connections and regions for classification.

The novelty lies in the integration of a deep neural network to estimate FC from lesion-derived SC (eSC) within the NeMo framework, creating a comprehensive pipeline for estimating both connectivity types from lesion masks alone.

Datasets

  • Patient Data: 100 individuals with multiple sclerosis (pwMS) were included.
  • 67 pwMS had no disability (EDSS < 2).
  • 33 pwMS had evidence of disability (EDSS ≥ 2).
  • Reference Data: Tractography results from 420 unrelated individuals from the Human Connectome Project were used as a reference for estimating SC and FC.

Key Insights and Ideas

  • Lesion-based estimation is powerful: The core finding is that connectivity estimated solely from lesion masks can be more effective for disability classification than directly measured connectivity from advanced MRI. This suggests that the impact of lesions on the network, rather than the precise state of the remaining connections, is a strong predictor of disability.
  • Regional estimates shine: Regional connectivity metrics (node strength) consistently outperformed pairwise connections for both observed and estimated networks, particularly for eSC and eFC. This implies that understanding the overall "health" or connectivity of specific brain regions is more informative than individual connections.
  • eFC is a strong contender: The regional eFC models achieved the highest classification accuracy, suggesting that the NeMo Tool's ability to estimate functional connectivity from lesion data is particularly promising.
  • Clinical feasibility: The NeMo Tool's reliance on conventional MRI and lesion masks makes it a much more accessible and cost-effective alternative to advanced MRI for large-scale studies and potentially for clinical applications.

Key Results

  • Outperformance of estimated models: Regional eSC and eFC models significantly outperformed their observed SC and FC counterparts (p < 0.05).
  • Highest AUC: Regional eFC models achieved the highest median AUC of 0.681 across all tested models.
  • Correlation of feature weights: Feature weights from regional eFC and observed FC models were moderately correlated (Pearson's r = 0.52, p < 10e-7), suggesting some shared underlying relationships, even with different data sources.
  • Disability-associated connectivity changes:
    • Increased estimated FC node strength in the visual network was associated with disability.
  • Decreased estimated SC node strength in the default mode and ventral attention networks was associated with disability.
  • Demographics and disability: Older age and being male were associated with a higher probability of having a disability.

Limitations and Future Work

  • Quality of advanced MRI: The study acknowledges limitations in the quality of the advanced MRI data collected from pwMS (shorter fMRI acquisition, single b-value for dMRI). This could have impacted the accuracy of the observed connectivity measures.
  • Reference population: The eSC and eFC are based on a population of 420 controls. This might not perfectly capture an individual's unique SC anatomy or how their specific connections are affected by lesions.
  • Compensatory rewiring: The NeMo Tool might not capture compensatory rewiring or reorganization of SC/FC that occurs in response to damage, which could be important for understanding recovery mechanisms.
  • Pathologies beyond lesions: The NeMo Tool only considers information from lesion masks, potentially missing other brain pathologies that could influence connectivity.
  • Coregistration accuracy: Lesion masks need to be coregistered to a common space, which can be challenging in cases of significant brain atrophy.
  • Future Work:
    • Compare eSC/eFC with SC/FC from higher-quality MRI acquisitions.
  • Investigate the prognostic ability of estimated connectomes by mapping baseline connectivity to future disability.
  • Further understand the role of the connectome in MS for developing better prognostic and therapeutic tools.

Summary updated

Abstract

Abstract Background Multiple Sclerosis (MS), a neurodegenerative and neuroinflammatory disease, causing lesions that disrupt the brain’s anatomical and physiological connectivity networks, resulting in cognitive, visual and/or motor disabilities. Advanced imaging techniques like diffusion and functional MRI allow measurement of the brain’s structural connectivity (SC) and functional connectivity (FC) networks, and can enable a better understanding of how their disruptions cause disability in people with MS (pwMS). However, advanced MRI techniques are used mainly for research purposes as they are expensive, time-consuming and require high-level expertise to acquire and process. As an alternative, the Network Modification (NeMo) Tool can be used to estimate SC and FC using lesion masks derived from pwMS and a reference set of controls’ connectivity networks. Objective Here, we test the hypothesis that estimated SC and FC (eSC and eFC) from the NeMo Tool, based only on an individual’s lesion masks, can be used to classify pwMS into disability categories just as well as SC and FC extracted from advanced MRI directly in pwMS. We also aim to find the connections most important for differentiating between no disability vs evidence of disability groups. Materials and Methods One hundred pwMS (age:45.5 ± 11.4 years, 66% female, disease duration: 12.97 8.07 years) were included in this study. Expanded Disability Status Scale (EDSS) was used to assess disability, 67 pwMS had no disability (EDSS &lt; 2). Observed SC and FC were extracted from diffusion and functional MRI directly in pwMS, respectively. The NeMo Tool was used to estimate the remaining structural connectome (eSC), by removing streamlines in a reference set of tractograms that intersected the lesion mask. The NeMo Tool’s eSC was used then as input to a deep neural network to estimate the corresponding FC (eFC). Logistic regression with ridge regularization was used to classify pwMS into disability categories (no disability vs evidence of disability), based on demographics/clinical information (sex, age, race, disease duration, clinical phenotype, and spinal lesion burden) and either pairwise entries or regional summaries from one of the following matrices: SC, FC, eSC, and eFC. The area under the ROC curve (AUC) was used to assess the classification performance. Both univariate statistics and parameter coefficients from the classification models were used to identify features important to differentiating between the groups. Results The regional eSC and eFC models outperformed their observed FC and SC counterparts (p-value &lt; 0.05), while the pairwise eSC and SC performed similarly (p=0.10). Regional eSC and eFC models had higher AUC (0.66-0.68) than the pairwise models (0.60-0.65), with regional eFC having highest classification accuracy across all models. Ridge regression coefficients for the regional eFC and regional observed FC models were significantly correlated (Pearson’s r = 0.52, p-value &lt; 10e-7). Decreased estimated SC node strength in default mode and ventral attention networks and increased eFC node strength in visual networks was associated with evidence of disability. Discussion Here, for the first time, we use clinically-acquired lesion masks to estimate both structural and functional connectomes in patient populations to better understand brain lesion-dysfunction mapping in pwMS. Models based on the NeMo Tool’s estimates of SC and FC better classified pwMS by disability level than SC and FC observed directly in the individual using advanced MRI. This work provides a viable alternative to performing high-cost, advanced MRI in patient populations, bringing the connectome one step closer to the clinic. Highlights We compared the accuracy of models based on observed functional connectivity (FC) and structural connectivity (SC) networks extracted from advanced MRI and estimated FC and SC networks derived using only lesion masks from conventional MRI in classifying people with multiple sclerosis (pwMS) into disability groups. Estimated SC and FC generally outperformed observed SC and FC in classifying pwMS into no disability vs evidence of disability groups, with regional estimated SC and FC having the best performance. Increased estimated FC node strength of regions in the visual network was associated with disability. Decreased estimated SC node strength of regions in the default mode and ventral attention networks was associated with disability. Despite their varied sources of origin, feature weights for the regional estimated FC and the regional observed FC classification models was significantly correlated (Pearson’s r = 0.52, p-value &lt; 10e-7).
BibTeX
@misc{tozlu2021estimated,
  author = {Ceren Tozlu and Keith Jamison and Zijin Gu and Susan A. Gauthier and Amy Kuceyeski},
  title = {{Estimated connectivity networks outperform observed connectivity networks when classifying people with multiple sclerosis into disability groups}},
  year = {2021},
  doi = {10.1101/2021.06.07.447376},
  howpublished = {bioRxiv},
}
Figure