Predicting lateralization of human language processing with non-language functional activity
Poster No:
1346
Submission Type:
Abstract Submission
Authors:
Haokun Li1, Gaolang Gong2
Institutions:
1State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern Institute for Brain Resea, Beijing, China, 2State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern Institute for Brain Resea, Beijing, Beijing
First Author:
Haokun Li
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, Beijing
State Key Laboratory of Cognitive Neuroscience and Learning & IDG/McGovern Institute for Brain Resea
Beijing, Beijing
Introduction:
While leftward lateralization of human language processing has been well documented at the population level, determining the direction and degree of language lateralization at the individual level is still a challenge. Previous empirical studies on language lateralization largely confined to functional activity of specific component of language processing per se [1]. To date, it remains unclear how the functional lateralization of other non-language functions relates to language lateralization. In this study, we applied machine learning approaches to explore whether and how the language lateralization could be individually predicted by functional lateralization of a set of non-language functions, using fMRI data from over 1000 healthy adults.
Methods:
A total of 1005 subjects (female/male: 534/471; age: 28.72±3.7 years) from the 'HCP1200' dataset were included in our analysis. All the subjects completed functional MRI scanning for language, emotion, gambling, relational, social, and working memory tasks [2]. The structural and functional images was pre-processed using the HCP pipelines [2,3].
A group-average activation map of the main contrast of language task (i.e., 'story VS. baseline') was used to locate the ROIs (Fig. 1A). The mean of positive Cohen's d values across the vertices of the entire cortex was calculated as the threshold, and the core language areas were then identified as the HCP-MMP parcels above the threshold (Fig. 1B) [4] [5].
For each individual, the laterality effect of language was defined as (left-right) of the mean Z values across the vertices within all core language areas in each hemisphere. For the main contrasts (i.e., 'explanatory variable of interest VS. baseline') of the other 5 non-language tasks, the functional asymmetry values were calculated as above for each pair of HCP-MMP parcel (180 in total). Therefore, each individual ended up with an asymmetry feature vector of non-language tasks with a length of 180*5. A ridge regression with a nested 5-fold cross-validation approach (5F-CV) was then applied to predict the language laterality effect with the asymmetry feature vector of non-language tasks [6] (Fig. 1C). A nonparametric permutation method was used to assess the significance of the Pearson correlation coefficient, mean absolute error (MAE), and feature weight.
A group-average activation map of the main contrast of language task (i.e., 'story VS. baseline') was used to locate the ROIs (Fig. 1A). The mean of positive Cohen's d values across the vertices of the entire cortex was calculated as the threshold, and the core language areas were then identified as the HCP-MMP parcels above the threshold (Fig. 1B) [4] [5].
For each individual, the laterality effect of language was defined as (left-right) of the mean Z values across the vertices within all core language areas in each hemisphere. For the main contrasts (i.e., 'explanatory variable of interest VS. baseline') of the other 5 non-language tasks, the functional asymmetry values were calculated as above for each pair of HCP-MMP parcel (180 in total). Therefore, each individual ended up with an asymmetry feature vector of non-language tasks with a length of 180*5. A ridge regression with a nested 5-fold cross-validation approach (5F-CV) was then applied to predict the language laterality effect with the asymmetry feature vector of non-language tasks [6] (Fig. 1C). A nonparametric permutation method was used to assess the significance of the Pearson correlation coefficient, mean absolute error (MAE), and feature weight.
Results:
As shown in Fig. 2A, the whole-brain functional lateralization of non-language functions could significantly predict language lateralization at the individual level (mean prediction accuracies r=0.543, p<0.0005, MAE=0.556, p<0.0005). In Fig. 2B, the absolute value of the weight represents the importance of corresponding feature in the prediction model, and the positive or negative feature value indicates whether the asymmetry feature changes in a similar way with language lateralization or the opposite. The significant or highly contributing non-language asymmetry features are always located within the core language areas across all non-language functions (the same trend for uncorrected gambling). Notably, the selected contrasts of these non-language tasks did not involve language processing, and therefore did not showed significant activation in these identified core language areas. Intriguingly, the significant asymmetry features are largely positive for the relational and working memory tasks, but negative for the other three non-language functions.
Conclusions:
The present study demonstrated that the language lateralization could be individually predicted by multivariate machine learning approach together with non-language functional lateralization features. This suggested a complex relationship of functional lateralization between language and non-language functions through core language areas.
Higher Cognitive Functions:
Higher Cognitive Functions Other
Language:
Language Comprehension and Semantics 2
Modeling and Analysis Methods:
Activation (eg. BOLD task-fMRI) 1
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
FUNCTIONAL MRI
Language
Machine Learning
1|2Indicates the priority used for review
Provide references using author date format
[2] Barch, D. M. et al. (2013), “Function in the human connectome: Task-fMRI and individual differences in behavior”, NeuroImage, vol. 80, pp. 169–189
[3] Glasser, M. F. et al. (2013), “The minimal preprocessing pipelines for the Human Connectome Project”, NeuroImage, vol. 80, pp. 105–124
[4] Glasser, M. F. et al. (2016), “A multi-modal parcellation of human cerebral cortex”, Nature, vol. 536, no. 7615, pp. 171–178
[5] Rajimehr, R. et al. (2022), “Complementary hemispheric lateralization of language and social processing in the human brain”, Cell Reports, vol. 41, no. 6, pp. 111617
[6] Cui, Z. et al. (2018), “The effect of machine learning regression algorithms and sample size on individualized behavioral prediction with functional connectivity features”, NeuroImage, vol. 178, pp. 622–637