hide
Free keywords:
-
Abstract:
Introduction
Low Frequency Fluctuations (LFFs) are known to represent a large portion of the variance of the BOLD signal. Many studies have claimed to identify activation distinct to superior temporal gyrus (STG) and superior temporal sulcus (STS) ([1],[2]) in language and hearing related processes. Using a data-driven clustering technique applied to LFFs, we investigated the spatial coherence and functional connectivity of the upper temporal lobe and sought to substantiate the suggested functional distinction of gyrus and sulcus.
Methods
Functional MRI/EPI data were acquired of 17 normal volunteers on a 3T MRI scanner (Siemens Trio) using TR=9 sec, TE=30ms, 3×3 mm⁁2 in-plane resolution, 3 mm slice thickness, 1mm gap between slices. The subjects performed a passive listening task and heard German sentences, both correct as well as syntactically violated, in intelligible and unintelligible format. All data sets were initially registered to an AC/CP coordinate system where the data were resampled to an isotropic voxel grid with a resolution of 3×3×3 mm⁁3. A general linear model was fitted to the data such that Y = × b + e where Y denotes a measured time course in one voxel, × denotes the design matrix and e the residuals. Our analysis focuses on the variance not explained by the experimental design contained in the residuals. The residuals e were bandpass-filtered so that only low-frequency components between 0.05 Hz and 0.01 Hz remained. We manually delineated an anatomical region of interest covering the entire STS/STG region. We then set up a similarity matrix where each entry r_ij denotes the correlation of the bandpass filtered residuals in voxel i and j. We transformed the correlation values r using Fisher's transform log((1+r)/(1-r)) and averaged the transformed correlation matrices across all test subjects. We then applied spectral clustering [3] to the averaged correlation matrix.
Results
Spectral clustering of residual low-frequency fluctuations in fMRI data showed a distinct separation between STG and STS (fig1a). Using inter-subject consistency as a criterion we found that 4 or 5 clusters best describe the subdivision. Common to both solutions is a clear-cut cluster separation that follows the anatomical separation between STS and STG (fig 1a). We then averaged time courses of the lowpass filtered residuals within each cluster of the 4-cluster map. These averaged time courses were subsequently used as covariates in a general linear model. The regression analysis showed that the STS is significantly more strongly correlated than the STG with the angular gyrus and the postcentral sulcus (fig 1b), whilst in contrast the STG shows significantly stronger correlations than the STS with the hippocampus and the intraparietal sulcus (fig 1c) with z>5.0 (uncorrected).
Conclusions
Spectral clustering of low-frequency fluctuations appears as a powerful tool to detect underlying commonalities in hemodynamic behaviour across a set of voxels. A subsequent regression analysis revealed distinctive correlational patterns associated with this subdivision whose interpretation will be the subject of future research.
