Multi-network Structural Connectome Predictors of Individual Language Learning Success
Poster No:
797
Submission Type:
Late-Breaking Abstract Submission
Authors:
Peilun Song1, Shuguang Yang2, Xiujuan Geng1, Suiping Wang2, Gangyi Feng1
Institutions:
1The Chinese University of Hong Kong, Hong Kong, NA, 2South China Normal University, GuangZhou, GuangDong
First Author:
Co-Author(s):
Introduction:
Human brains vary in language learning ability, traditionally linked to specific brain regions. However, recent evidence suggests that multiple neural networks, including the default mode network (DMN), fronto-parietal network (FPN), dorsal attention network (DAN), and orbito-affective network, interact with language-related regions to support language learning. This study explores whether structural connectivity, measured through diffusion tensor imaging (DTI) and structural MRI (sMRI), can predict individual differences in language learning and identifies neuromarkers associated with general and specific language learning abilities.
Methods:
102 healthy adults (72 females; age=21.29±2.03 years) was enrolled in a 7-day artificial language training program. The training included six learning tasks: morphology learning (NP), phrase structure learning (SV), sentence structure learning (SOV), sound category learning (RCat), vowel category learning (VCat), and word learning (Word). PCA was applied to the 7-day learning scores to disentangle general and specific components of language learning ability.
CAT12 toolbox was used to process the T1-weighted data for voxel-based morphometry and parcellated into 360 regions of interest (ROIs) using the HCP atlas (Glasser et al., 2016). Forty-three radiomics features were calculated for each ROI, and inter-regional Pearson correlations produced a morphological connectome matrix (Liu et al., 2024).
DTI data were processed using PANDA to calculate fractional anisotropy (FA), and deterministic fiber tracking was performed. Structural connectivity matrices were constructed using average FA values of fibers connecting each pair of regions.
To characterize brain topological properties, nine node-wise graph theory metrics were calculated: Betweenness Centrality, Degree Centrality, Node Cluster Coefficient, Node Efficiency, Node Local Efficiency, Page Rank Centrality, Eigenvector Centrality, Participant Coefficient, and Subgraph Centrality.
For prediction, we applied random forest regression along with lasso feature selection based on a nested five-fold cross-validation. We calculated Pearson correlations to assess model performance. This procedure was repeated 1,000 times to ensure robustness. Our evaluation considered predictions of general language learning ability (PC1) and combinations with specific components (PC1+2 through PC1+6).
CAT12 toolbox was used to process the T1-weighted data for voxel-based morphometry and parcellated into 360 regions of interest (ROIs) using the HCP atlas (Glasser et al., 2016). Forty-three radiomics features were calculated for each ROI, and inter-regional Pearson correlations produced a morphological connectome matrix (Liu et al., 2024).
DTI data were processed using PANDA to calculate fractional anisotropy (FA), and deterministic fiber tracking was performed. Structural connectivity matrices were constructed using average FA values of fibers connecting each pair of regions.
To characterize brain topological properties, nine node-wise graph theory metrics were calculated: Betweenness Centrality, Degree Centrality, Node Cluster Coefficient, Node Efficiency, Node Local Efficiency, Page Rank Centrality, Eigenvector Centrality, Participant Coefficient, and Subgraph Centrality.
For prediction, we applied random forest regression along with lasso feature selection based on a nested five-fold cross-validation. We calculated Pearson correlations to assess model performance. This procedure was repeated 1,000 times to ensure robustness. Our evaluation considered predictions of general language learning ability (PC1) and combinations with specific components (PC1+2 through PC1+6).
Results:
PCA of language learning scores revealed that PC1-PC3 showed significant learning patterns over the seven days, while PC4-PC6 remained relatively stable. As shown in Fig.1, PC1 represented general language learning trajectories with relatively equal contributions from all tasks. PC2, PC3, PC4, PC5, and PC6 reflected specific learning components dominated by particular tasks.
Prediction analyses (Fig.2) demonstrated that using all features could significantly predict all learning outcomes. Notably, the measures from the orbito-affective network (OAN) robustly predicted all learning outcomes. Additionally, features from the DAN, ventral-multimodal networks predicted PC1+2 with performance comparable to using all features. PC1+3 and PC1+4 could be predicted using single-network features (posterior-multimodal and somatomotor networks).
Analysis of feature importance (Fig.3) revealed that degree centrality of the OAN was the most predictive metric across all six prediction tasks. For PC1+2, PC1+3, and PC1+4, Eigenvector centrality and subgraph centrality were also important.
Prediction analyses (Fig.2) demonstrated that using all features could significantly predict all learning outcomes. Notably, the measures from the orbito-affective network (OAN) robustly predicted all learning outcomes. Additionally, features from the DAN, ventral-multimodal networks predicted PC1+2 with performance comparable to using all features. PC1+3 and PC1+4 could be predicted using single-network features (posterior-multimodal and somatomotor networks).
Analysis of feature importance (Fig.3) revealed that degree centrality of the OAN was the most predictive metric across all six prediction tasks. For PC1+2, PC1+3, and PC1+4, Eigenvector centrality and subgraph centrality were also important.
Conclusions:
Results show that structural connectomes, especially from DTI, can predict language learning outcomes. This supports the multiple neural network hypothesis, highlighting the importance of connectivity across domain-general networks. Degree centrality of the OAN was a key predictor, emphasizing the role of reward and motivation circuits. The DMN and FPN also showed consistent connections with the reward network, aligning with theories that language learning relies on multiple interacting neural systems.
Language:
Language Acquisition 1
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Connectivity (eg. functional, effective, structural)
Diffusion MRI Modeling and Analysis
Multivariate Approaches
Keywords:
Other - Language Learning, Individual Difference, Multimodal MRI, Predictive Modeling, Graph Theory
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
Provide references using APA citation style.
Liu H, Ma Z, Wei L, et al. A radiomics-based brain network in T1 images: Construction, attributes, and applications. Cerebral Cortex. 2024; 34(2):bhae016.