Objective: Using linear discriminant analysis (i.e., canonical variate analysis (CVA)), we present preliminary results decomposing columnar-resolution, cerebral blood volume (CBV)-weighted fMRI signals in the visual cortex of the cat into three significant components following visual stimulation with 0° and 90° drifting gratings.

Figure 1.


Figure 2.

Methods: Columnar-resolution, CBV-weighted fMRI signals were obtained in a 1-mm thick slice tangential to the surface of the cortex containing visual area 18, using gradient-echo data collection (in-plane resolution=0.15 x 0.15 mm², TE=10ms and TR=2s) at 9.4 Tesla after injection of MION contrast in an isoflurane- anesthetized cat [1]. Stimuli consisted of square-wave, high-contrast, moving gratings with low spatial frequency at two orthogonal orientations (0° vs. 90°). Each epoch consisted of 10 baseline, 10 stimulus, and 9 baseline scans—baselines contained stationary grating patterns with the same orientation—for 29 time points per epoch. Interleaved 0° and 90° epochs were repeated 40 times, each with a ~30s break between epochs. Using the NPAIRS software package, after an initial principal component analysis 100 principal components were retained and epoch means were removed with GLM preprocessing. Within NPAIRS the 40 interleaved epochs were split, 50 times, into two independent sets of 20 interleaved epochs. For each set of 20 interleaved epochs CVA was applied with 58 discriminant groups, one for each time point (scan) in the 0° (29 scans) and 90° (29 scans) epochs. This approach extracts a multidimensional, model-free temporal response for 0° and 90° epochs under the assumption of stationarity across the 20 interleaved epochs. By comparing eigenimages (EIs) across split-half sets we extracted Z-scored SPMs and reproducibility metrics for the 57 dimensions obtained from 58 groups[2].

Results & Discussion: Figure 1 demonstrates canonical variate scores (CVSs) as a function of group number (time/epoch) for the first 4 dimensions, and Figure 2 the corresponding Z-scored EIs (upper-left = EI1, lower-right = EI4). Dimensions 1-4 accounted for 51.6%, 22.4%, 5.8% and 1.3% of the total variance, with EI reproducibilities of 0.5, 0.4, 0.25, and 0.05, respectively. EI1 reflects a general cortical response with hemodynamic response functions (HRFs) in CVS1 independent of grating orientation. CVS2 reflects the differential columnar response to the stimulation at each orientation, which exists on top of CVS1, with complementary spatial patches in the visual cortex area(EI2:0°=green; 90°=yellow). CVS3 reflects an approximate first derivative of the CVS1 time course, possibly due to surface and middle line areas with a faster rise in their HRFs. EI4/CVS4 is likely noise given its low reproducibility index (~0.05).

Conclusions: NPAIRS/CVA demonstrates a powerful capability to extract multiple hemodynamic components in the analysis of columnar-resolution, CBV-weighted fMRI from the cat's visual cortex.

References & Acknowledgements:
[1] Zhao F, et al., Spatial Specificity of Cerebral Blood Volume-weighted fMRI Responses at Columnar Resolution (submitted to NeuroImage).
[2] Strother SC, et al., NeuroImage 15:747-771, 2002.

This work was in part supported by NIH grants EB002013 and EB003324.