Latent connectivity-behavior dimensions following stroke
Poster No:
1100
Submission Type:
Abstract Submission
Authors:
Ke Wu1, Gong Gaolang1
Institutions:
1State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern,Beijing Normal University, Beijing, China
First Author:
Ke Wu
State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern,Beijing Normal University
Beijing, China
State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern,Beijing Normal University
Beijing, China
Co-Author:
Gong Gaolang
State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern,Beijing Normal University
Beijing, China
State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern,Beijing Normal University
Beijing, China
Introduction:
Understanding neurobiological mechanisms behind behavioral deficits is crucial for prognosing and treating stroke patients [1, 2]. To address this issue, previous neuroimaging studies on stroke patients largely applied univariate correlation or machine learning approaches to predict each single behavioral score with single or multiple brain features [3, 4]. These approaches, however, ignore the highly correlation nature of post-stroke deficits across behavioral domains and possible latent dimensions of heterogeneous behavioral phenotypes in stroke patients [4, 5]. The present study applied a multivariate data-driven method to explore the latent dimensional associations between post-stroke behavioral measures of multiple domains and distributed functional connectivity of the entire brain.
Methods:
Dataset
A public shared longitudinal stroke dataset from the Corbetta's group was used [1, 2, 4]. This dataset includes a battery of neuropsychological tests and multi-modal MRI scanning at three poststroke time points: ~ 2 weeks, ~3 months, and ~12 months. For each time point, the patients with all behavioral assessments and qualified resting-state functional MRI scans were entered into our analysis (~2 weeks: 57 patients; ~3 months: 63 patients; ~12 months: 63 patients; 31 healthy controls).
Data processing
For each of the 5 neuropsychological domains (motor, language, spatial attention, verbal memory, and spatial memory), principal component analysis (PCA) was applied to all within-domain tests, resulting in 5 component scores.
For each patient, stroke lesions have been manually outlined by the Corbetta's group. Resting-state functional data was minimally preprocessed using FSL [4]. We then applied the Gordon 333 cortical parcellation and Pearson correlation to estimate a 333×333 functional connectivity (FC) matrix [5]. Here, lesion voxels within each parcel were masked out, and the functional connectivity from the parcel with more than 50% lesion voxels were excluded from the FC matrix [6].
Statistical analysis
The partial least squares correlation (PLSC) was applied to identify significant latent components (LC) between the multi-domain neurological scores and FC measures within the matrix.
Each identified LC accompanies with a whole-brain weighted matrix pattern (i.e., FC salience) and behavioral profile (i.e., behavioral salience), and a FC composite score and a behavioral composite score then can be yielded for each individual. Pearson correlations between the original data and resultant LC composite scores was used to quantify the contribution of each raw variable to the LCs, referred to as LC loading.
A public shared longitudinal stroke dataset from the Corbetta's group was used [1, 2, 4]. This dataset includes a battery of neuropsychological tests and multi-modal MRI scanning at three poststroke time points: ~ 2 weeks, ~3 months, and ~12 months. For each time point, the patients with all behavioral assessments and qualified resting-state functional MRI scans were entered into our analysis (~2 weeks: 57 patients; ~3 months: 63 patients; ~12 months: 63 patients; 31 healthy controls).
Data processing
For each of the 5 neuropsychological domains (motor, language, spatial attention, verbal memory, and spatial memory), principal component analysis (PCA) was applied to all within-domain tests, resulting in 5 component scores.
For each patient, stroke lesions have been manually outlined by the Corbetta's group. Resting-state functional data was minimally preprocessed using FSL [4]. We then applied the Gordon 333 cortical parcellation and Pearson correlation to estimate a 333×333 functional connectivity (FC) matrix [5]. Here, lesion voxels within each parcel were masked out, and the functional connectivity from the parcel with more than 50% lesion voxels were excluded from the FC matrix [6].
Statistical analysis
The partial least squares correlation (PLSC) was applied to identify significant latent components (LC) between the multi-domain neurological scores and FC measures within the matrix.
Each identified LC accompanies with a whole-brain weighted matrix pattern (i.e., FC salience) and behavioral profile (i.e., behavioral salience), and a FC composite score and a behavioral composite score then can be yielded for each individual. Pearson correlations between the original data and resultant LC composite scores was used to quantify the contribution of each raw variable to the LCs, referred to as LC loading.
Results:
As shown in Figure 1, PLSC analysis consistently revealed one significant LC for poststroke three time points, as well as for the healthy controls. The loadings of the original features (i.e., 5 behavioral domain scores and 55278 functional connectivity) were illustrated in Figure 1B-C.
Interestingly, both FC saliences and loadings exhibited overall similar pattern among the three poststroke time points by showing significant correlations between each other (Fig.2). However, their correlation with health controls was gradually increased along the poststroke time, suggesting a reorganization of the behavior-connectivity relationship following the gradual poststroke recovery.
Interestingly, both FC saliences and loadings exhibited overall similar pattern among the three poststroke time points by showing significant correlations between each other (Fig.2). However, their correlation with health controls was gradually increased along the poststroke time, suggesting a reorganization of the behavior-connectivity relationship following the gradual poststroke recovery.
Conclusions:
Using a multivariate data-driven approach, the present study revealed a robust latent connectivity-behavior component/association in both stroke patients and healthy controls. Particularly, the poststroke FC pattern underlying such association is gradually changed toward the pattern in healthy controls, supporting a dynamic reorganization of functional connectivity networks after the stroke.
Learning and Memory:
Neural Plasticity and Recovery of Function 1
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural)
Multivariate Approaches 2
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
FUNCTIONAL MRI
Language
Motor
Multivariate
Open Data
Other - Stroke
1|2Indicates the priority used for review
Provide references using author date format
[2] Ramsey, L. E.et al. (2017), 'Behavioural clusters and predictors of performance during recovery from stroke', Nature human behaviour, vol. 1, no. 3, pp. 0038.
[3] Baldassarre, A.et al. (2014). 'Large-scale changes in network interactions as a physiological signature of spatial neglect'. Brain, vol. 137, no. 12, pp. 3267-3283.
[4] Siegel, J. S. et al. (2016), 'Disruptions of network connectivity predict impairment in multiple behavioral domains after stroke', Proceedings of the National Academy of Sciences, vol. 113, no. 30, pp. E4367-E4376.
[5] Genon, S. et al. (2022), 'Linking interindividual variability in brain structure to behaviour', Nature Reviews Neuroscience, vol. 23, no. 5, pp. 307-318.
[6] Gordon EM, et al. (2016), 'Generation and evaluation of a cortical area parcellation from resting-state correlations', Cerebral Cortex, vol. 26, no. 1, pp. 288–303
[7] Siegel, J. S. et al. (2018). 'Re-emergence of modular brain networks in stroke recovery', Cortex, vol. 101, pp. 44-59.