Functional connectivity predicts speech perception activation around the primary auditory cortex
Poster No:
1755
Submission Type:
Abstract Submission
Authors:
Junhao Luo1, Gaolang Gong1
Institutions:
1State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern Institute for Brain Resea, Beijing, China
First Author:
Junhao Luo
State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern Institute for Brain Resea
Beijing, China
State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern Institute for Brain Resea
Beijing, China
Co-Author:
Gaolang Gong
State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern Institute for Brain Resea
Beijing, China
State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern Institute for Brain Resea
Beijing, China
Introduction:
The relationship between connectivity patterns and brain functions is crucial in neuroscience. Previous studies have demonstrated that connectivity patterns can predict functional activation in various brain regions, with considerable individual differences in prediction scores [1], [2]. However, such individual difference and its neurobiological underpinnings are largely overlooked. This study applied the resting-state functional connectivity (rsFC) from the primary auditory cortex (PAC) to predict its speech perception activation and further explored the individual difference of the prediction score.
Methods:
Participants. Following quality assurance, 766 subjects with high-quality magnetic resonance imaging (MRI) from the Human Connectome Project (HCP) dataset were included.
The defined PAC. For each subject, the PAC is determined as the combination of the planum temporale (PT) and Heschl's gyrus (HG). These two structures were manually delineated by two trained raters.
Vertex-level rsFC. For each vertex on the PAC, 1\times400 rsFC vector was obtained with each element as the Pearson's correlation between its resting-state time series and the averaged time series of a parcel from the "Kong400" parcellation [3].
Functional activation. Four task functional MRI scans from the HCP S1200 dataset were utilized, including language, gambling, social, and emotion. The activation was represented by the t-value. Given the high correlation, the "story - baseline" and "math - baseline" contrasts of the language task were averaged, representing the speech perception. The "story - math" contrast was derived to represent speech comprehension [4].
Predictive framework. The ridge regression algorithm with a nested five-fold cross-validation was applied (Fig. 1). In addition, an independent dataset [5] was used to further evaluate model generalizability.
Prediction score (PS). For each subject, the prediction score was defined as the Pearson's correlation between the actual and predicted activation across all PAC vertices.
Statistical analyses. A general linear model was applied to investigate the associations of prediction score with structural measures and functional activation with the PAC, while controlling for age, sex, and PAC vertex number.
The defined PAC. For each subject, the PAC is determined as the combination of the planum temporale (PT) and Heschl's gyrus (HG). These two structures were manually delineated by two trained raters.
Vertex-level rsFC. For each vertex on the PAC, 1\times400 rsFC vector was obtained with each element as the Pearson's correlation between its resting-state time series and the averaged time series of a parcel from the "Kong400" parcellation [3].
Functional activation. Four task functional MRI scans from the HCP S1200 dataset were utilized, including language, gambling, social, and emotion. The activation was represented by the t-value. Given the high correlation, the "story - baseline" and "math - baseline" contrasts of the language task were averaged, representing the speech perception. The "story - math" contrast was derived to represent speech comprehension [4].
Predictive framework. The ridge regression algorithm with a nested five-fold cross-validation was applied (Fig. 1). In addition, an independent dataset [5] was used to further evaluate model generalizability.
Prediction score (PS). For each subject, the prediction score was defined as the Pearson's correlation between the actual and predicted activation across all PAC vertices.
Statistical analyses. A general linear model was applied to investigate the associations of prediction score with structural measures and functional activation with the PAC, while controlling for age, sex, and PAC vertex number.
Results:
As shown in Fig. 2A, the predicted PAC activation pattern from the rsFC highly resembles the actual activation pattern in selected subjects. Particularly, the model trained from all HCP subjects can be applied to well predict the activation of speech perception for most subjects from an independent validation dataset, indicating robust generalization (Fig. 2C-D).
For both speech perception and comprehension, there exist substantial individual difference in the prediction score for both left and right PAC across HCP subjects. As shown, the prediction scores of speech perception are significantly higher than the ones from both speech comprehension and other fMRI contrasts, suggesting a greater and specific rsFC predictive capability to the activation of speech perception within the PAC (Fig. 2B).
Finally, the prediction scores for speech perception only showed a significant positive correlation with both peak value and variation of functional activation within the PAC (Fig. 2E, All related r > 0.3, p < 2.2×10-16), but no correlation with any of the PAC structural measures (Fig. 2E). In addition, there was a significant correlation between the prediction score of the left PAC and reading decoding score (r = 0.11, p = 0.003).
For both speech perception and comprehension, there exist substantial individual difference in the prediction score for both left and right PAC across HCP subjects. As shown, the prediction scores of speech perception are significantly higher than the ones from both speech comprehension and other fMRI contrasts, suggesting a greater and specific rsFC predictive capability to the activation of speech perception within the PAC (Fig. 2B).
Finally, the prediction scores for speech perception only showed a significant positive correlation with both peak value and variation of functional activation within the PAC (Fig. 2E, All related r > 0.3, p < 2.2×10-16), but no correlation with any of the PAC structural measures (Fig. 2E). In addition, there was a significant correlation between the prediction score of the left PAC and reading decoding score (r = 0.11, p = 0.003).
Conclusions:
Our study demonstrated a robust and specific rsFC-based model of individually predicting the activation pattern of speech perception within the PAC. This predictive model showed individual difference in prediction performance, which correlates with specific individual and therefore is biologically meaningful.
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 1
Perception, Attention and Motor Behavior:
Perception: Auditory/ Vestibular 2
Keywords:
Other - resting-state functional connectivity (rsFC), functional activation, primary auditory cortex (PAC), prediction scores, speech perception, individual difference
1|2Indicates the priority used for review
Provide references using author date format
[2] Saygin, Z. M. et al. (2012), “Anatomical connectivity patterns predict face selectivity in the fusiform gyrus”, Nature Neuroscience, vol. 15, no. 2, pp. 321-327
[3] Kong, R. et al. (2021), “Individual-Specific Areal-Level Parcellations Improve Functional Connectivity Prediction of Behavior”, Cerebral Cortex, vol. 31, no. 10, pp. 4477–4500
[4] Binder, J. R. et al. (2011), “Mapping Anterior Temporal Lobe Language Areas with FMRI: A Multi-Center Normative Study”, NeuroImage, vol. 54, no. 2, pp. 1465–1475
[5] Wang, S. et al. (2022), “A synchronized multimodal neuroimaging dataset for studying brain language processing”, Scientific Data, vol. 9, no. 1, pp. 590
[6] Barch D. M. et al. (2013), “Function in the human connectome: task-fMRI and individual differences in behavior”, NeuroImage, vol. 80, pp. 169–189