Multi-node directed cortical network for speech processing revealed by multivariate Granger causality analysis
Yayan Yin1, Jiahong Gao1, Bing Wu2, Yang Fan2, Bingjiang lyu1, and Jianqiao Ge1

1Peking University, Beijing, China, People's Republic of, 2GE Healthcare, Beijing, China, People's Republic of

Synopsis

For decades, how the information flows among multiple brain regions remains unclear for speech processing, due to the challenge of mapping multi-node directed cortical pathways from brain images. In this work, multivariate Granger causality analysis is employed on functional MR images to reveal the effective connectivity of Chinese language-speech network for the first time. The results showed that left insula and posterior middle temporal gyrus were the strong driver nodes, the left middle frontal gyrus and superior temporal gyrus were the most received nodes in the network. We also found greater interhemispheric connectivity in females compared to males.

Purpose

The question of how the brain processes for intelligible speech in auditory stimulation is a significant problem still unsolved. It is not enough just analyze directed interactions among part of main activity regions. In the present study, we used the multivariate Granger causality analysis and graph theory to investigate Chinese intelligible speech network. Attention we also paid on the gender difference on these cortical connections.

Methods

Twenty-eight Chinese native speakers (15 males and 13 females, aged between 21 and 28, mean age 24.2 years) were recruited for this study, consent forms were obtained prior to the scan. The participants were presented with alternating intelligible and unintelligible Mandarin Chinese speech blocks, while BOLD acquisitions were made on a whole body 3.0T scanner equipped with a head coil. Thirty-five transversal slices of functional images that covered the whole brain were acquired using a gradient-echo echo-planar imaging (EPI) pulse sequence (TR/TE/θ = 2.08 s/30 ms/90º, 64 x 64 x 35 matrix with 3 x 3 x 3 mm3 spatial resolution). Four sessions of functional task scanning were acquired and each session started with a blank screen for 10s, then followed by nine blocks of auditory stimuli, lasted 378.56s in total. High resolution anatomical images were obtained using a 3D T1-weighted MPRAGE sequence (TR/TE = 2.6 s/3.02 ms, 224 x 256 x 176 matrix with 1 x 1 x 1 mm3 spatial resolution). Participants were instructed to only judge the gender of the speech played in both intelligible and unintelligible language blocks. Twelve brain regions were first identified in the whole-brain volume using SPM analysis based on contrast in intelligible acquisition greater than that of unintelligible acquisition. The Granger causality analysis1 was then conducted using a variation of direct directed transfer function (dDTF) approach on the fMRI data, and computed from a multivariate autoregressive model of the times series in the identified ROIs. The raw time series in the all ROIs were normalized across runs and subjects, and then all the normalized time series from all runs and subjects were concatenated to form a single vector per ROI for analysis. Based on the derived dDTF causality map, the clustering coefficients of the network graph were also calculated2.

Results

The dDTF values (P<0.05) that reflect the level of causal influence among the 12 identified ROIs are shown in the check board table in Fig.1a, it can be seen that the influence is highest from the left insula (L.Ins) to left middle frontal gyrus (L.MFG) and also from the left posterior superior temporal gyrus (L.pSTG) to left anterior superior temporal gyrus (L.aSTG). The information flow between different nodes can be visualized in the clustering coefficient map shown in Fig.1b, where the color scale in inner ring and outer ring represents the clustering in and clustering out coefficients respectively. The results showed that L.Ins and L.pSTG were the main nodes driving other ROIs in the processing of intelligible speech, while L.MFG and the L.aSTG were the major information inflow nodes in the intelligible network. The dDTF values (P<0.05) for regions where male and female demonstrated higher level of connectivity are shown in Fig.2 left and right respectively. It can be seen that female had better right-to-left interhemispheric interaction, especially the right fusiform gyrus (R.FG) to L.MFG and L.FG; whereas male demonstrated more significant intrahemipheric in­­­­teractions in the left hemisphere which included the insula, the fr­­ontal lobes and fusiform gyrus.

Discussion and conclusion

In this work, multivariate Granger causality analysis was used to identify the connectivity network of intellectual speech processing based on fMRI acquisition. Granger analysis was chosen as it requires no prior knowledge of the underlying network and is computational light that may be applied to whole brain volume. Twelve brain regions were identified as nodes in the network and the major inflow and outflow nodes were identified. Difference in the connectivity network for male and female was also investigated that may provide evidence for further research on the gender differences of cortical effective connectivity.

Acknowledgements

No acknowledgement found.

References

1: Deshpande G, et al. Multivariate Granger causality analysis of fMRI data. Hum. Brain. Mapp. 2009; 30(4):1361-1373. 2: Watts DJ, Strogatz SH. Collective dynamics of ‘small-word’ network. Nature. 1998; 393(6684):440-442.

Figures

Fig.1. Effective connectivity network are displayed between 12 ROIs, (a) checkerboard network and (b) out ring is the clustering-out coefficients and inner ring is the clustering-in coefficients. Only displayed the significant paths (P<0.05), the influence direction as indicated by the white arrow, is from the columns to the rows.

Fig.2. Multivariate Granger causality analysis for gender differences. Left, male>female; Right, female>male. Only displayed the significant paths (P<0.05), the influence direction as indicated by the white arrow, is from the columns to the rows.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1701