An undersampled radial acquisition with 24 rays per image and a single saturation pulse for five slices was reconstructed with STCR to obtain increased coverage of the heart for myocardial perfusion. In this work we chose to increase the coverage by acquiring more short-axis slices but this can be traded for higher resolution (though resolution along each k-space ray is relatively high) or obtaining more slices with different orientations. The proposed acquisition results in different slices having different signal intensities and contrasts for a given time frame due to different saturation recovery times as implied by . The first slice that is acquired immediately after each saturation pulse, for example slice #1 and slice #2 in , have relatively poor image quality as compared to the other slices due to the low signal and more inconsistent projection data associated with their short saturation recovery times. These slices could be discarded or might be used to obtain a quantitative arterial input function (AIF) from the left ventricular blood pool. AIF from these short SRT slices is likely not saturated and has been used to obtain more accurate semi-quantitative and quantitative parameters (1
). The remaining eight slices can be used for both visual and quantitative analyses.
SRTs used here ranged from approximately 35–300 msec, with 90–300 msec being used in the visual assessment. Too long of a SRT could give inadequate T1-weightings. The range of SRTs used successfully in (1
), from 42–410 msec, implies these SRTs are not too long.
Of all the slices from 7 patients, slices from two patients had the best image quality (ex. Radial slices in – Patient #3) and slices from the other five patients had slightly lower image quality (ex. ). The positioning and the choice of which coil data are used to combine in SoS fashion from multiple coils plays an important role in the reconstructed image quality as some of the coils may not have any signal from the heart and only contribute to increased streaking. This type of coil data can be discarded from the SoS combination. Here the coil reconstructions which only contributed to streaking were manually (based on visual inspection of signal present in the heart region after reconstruction) excluded from the SoS combination.
From we see that the normalized reconstruction error decreases with increasing Gd concentration and flip angle. This is interesting as one might guess that with increasing [Gd] the signal changes more rapidly and hence would lead to increased error (as the reconstruction error is related to inconsistencies between radial rays in k-space). But with the presence of even a small time delay (TD) between the saturation pulse and the first readout α pulse, signal inconsistencies are smaller than might be expected. shows the changing signal intensities for the blood pool region in the simulated heart image as 24 rays in k-space are acquired for the first slice and for different Gd concentrations. … The signal is increasing as the concentration is increased and also as radial rays in k-space are acquired. But when the signal curves for different Gd concentrations are normalized so that all the curves start from the same value as shown in , we see that the signal variation is decreasing with increasing Gd concentration. To quantify the amount of variation in the normalized curves, gradient of each curve was computed using finite differences and its L1 norm was computed. The L1 norms of the gradients for the curves are shown next to each curve in . The gradients are decreased with increasing [Gd]. Although not shown here, a similar trend in the variation of the normalized signal was observed with increasing the readout flip angle for a given Gd concentration. The source of this error in the reconstruction is only due to non-uniformity in k-space as these simulations are noise free.
Figure 6 Signal intensity variation with changing [Gd]. (a) Plot comparing the signal intensities for different [Gd] for 24 readout lines. (b) Plot comparing the normalized signal intensities for different [Gd] for 24 readout lines. The number next to each [Gd] (more ...)
In this study, the patients were instructed to breathe shallowly as opposed to holding their breath during the scan. Holding their breath for part of the scan often leads to much deeper breath later on resulting in large sudden changes in time. The radial acquisition scheme and reconstruction method were found to be much more robust to modest respiratory motion than similar undersampled Cartesian STCR methods as shown in . This robustness to respiratory motion is an important finding and is largely due to the radial sampling scheme in which the center of k-space, where there is most energy, is densely sampled in multiple directions and so the fidelity term helps in preserving sudden changes in time due to respiratory motion. Also, using an L1 as the norm in the temporal constraint helps in better preserving sudden changes in time due to motion as compared to using an L2 norm. When there are no sudden changes in time (no motion), using the L2 norm gave results nearly identical to those obtained using an L1 norm (26). Using a spatial constraint along with the temporal constraint further helped to reduce streaking and resulted in better reconstructions. This is different from the well known result of robustness of radial sampling scheme to motion artifacts that are caused due to motion of the object as k-space lines are acquired.
The STCR reconstruction method here was applied to each coil data separately and the resulting images were combined in a square root of sum of squares fashion. In preliminary work, we have found small improvements by using joint multi-coil reconstruction techniques that use coil sensitivity information in the constrained reconstruction.
In conclusion, undersampled radial acquisition with a spatiotemporal constrained reconstruction method can be used to improve slice coverage of the heart for perfusion imaging. The radial STCR method produced images with better or comparable quality to standard Cartesian acquisitions and also with increased coverage. The method is relatively robust to respiratory motion in the data and results in images with high SNR and CNR. More work is warranted to test the method in patients with ischemia during vasodilation.