Figure 1.
Cartesian (A) and density weighted acquisition (B) for a typical fMRI experiment.
The MTF (bottom, grey) results from a multiplication of the k-space density ρ(k) (blue, top) with the signal S(k) (green) and the filter f(k) (red).
Figure 2.
Flow chart of the steps involved in the reconstruction of the fMRI timeseries.
The interleaved Cartesian and density weighted datasets are first splitted. Subsequently, unfiltered and filtered images are reconstructed for both acquisition methods, respectively. The filtered images are utilized for statistical processing, whereas the unfiltered images are not SNR efficient and solely created for more accurate motion correction and registration of the filtered data to the anatomical images.
Figure 3.
Phantom images and corresponding spatial response functions.
Images were reconstructed from k-space density weighted (A), Cartesian (B) and unfiltered Cartesian acquisition (C). Spatial response functions (D) were obtained from the edge spread functions indicated by the red bars.
Figure 4.
Representative slices of the brain of a healthy volunteer.
Images are shown for k-space density weighted (A), Cartesian filtered (B), unfiltered density weighted (C) and unfiltered Cartesian reconstructions (D). Cartesian and density weighted images correspond well in geometry and contrast.
Figure 5.
Bland-Altman difference plots of quantitative parameter gains.
Gains are plotted on a voxel-to-voxel basis for spatial SNR, temporal SNR and relative signal change (rSC; scaled to the mean Cartesian value) for density weighted (DW) over Cartesian (Cart) EPI in n = 5 subjects. Red solid lines represent the mean difference across voxels and subjects, red dashed lines ±1.96 × the standard deviation (SD; 95% limits of agreement for each comparison). The dotted gray line represents identity (no difference). Average increases in spatial SNR (12.4%, t = 15.68), temporal SNR (5.5%, t = 2.46) and rSC (8.6%, t = 2.63) were consistent and statistically significant (p<0.03125; based on mean within-subject differences). Gray data points and corresponding gray lines in the tSNR plot represent values of one additional subject measured at rest for comparison.
Figure 6.
Five consecutive slices of the statistical activation images thresholded using clusters (determined by Z>2.3 and a FWER-corrected p≤0.05) of the subject presented in Fig. 4 for the density weighted and the Cartesian acquisition.
Figure 7.
Second-level fixed-effects (FE) fMRI results - Mean activation.
Evoked by left-hand finger tapping in n = 5 subjects as detected by density weighted (top) and Cartesian (bottom) EPI acquisitions (all thresholded using clusters determined by Z>2.3 at a FWER-corrected p≤0.05 and projected to the pial surface of the MNI152 template).
Table 1.
Corresponding local activation maxima from second level fixed-effects (FE) analyses of the mean activation evoked by left-hand finger tapping as detected by density weighted and Cartesian EPI in n = 5 subjects.
Figure 8.
Second-level fixed-effects (FE) fMRI results - Differential contrast.
(A) Revealing a cluster of significantly increased activation detected by k-space density weighted compared to conventional Cartesian EPI for left-hand finger tapping in n = 5 subjects (thresholded using clusters determined by Z>2.3 at a FWER-corrected p≤0.05 and displayed in MNI152 standard space). In the opposite, no areas of increased activation detected by Cartesian over density weighted EPI were found. (B) Time-courses within this cluster (extracted from raw data prior to further processing and averaged across n = 5 subjects) reveal increased percentual BOLD signal changes of density weighted compared to Cartesian EPI for each of the five blocks of the finger-tapping task.