14 Sparse Sampling in MRI |
337 |
Fig. 14.9 The application of PECS to CE-MRA is illustrated. Samples in k-space samples are acquired progressively by means of different randomly selected subsets (top row). A combination of a set of the acquisitions is used to achieve a “reference” image which is sorted to derive R. PECS is applied to each of the (high acceleration factor) subsets using R. The result (bottom right) is a sequence of relatively high time resolution images. The arrow indicates a region in one of the output frames, which does not show an artifact which appears in the reference image
more detailed study. In each of the two lower rows are three different reconstructions obtained from a fraction of the k-space data, simulating the acceleration factors indicated. In the left column are shown the SENSE reconstructions, which are clearly noisy and unlikely to be diagnostically useful. CS reconstructions are shown in the center column; in this case, the sampling pattern employed was designed specifically for CS. The results are superior to the SENSE reconstructions, but somewhat blurred in appearance. SENSECS reconstructions are shown in the right column; they show better fidelity than the other reconstructions and diagnostically useful results up to at least an acceleration factor of 6.5. Note that the images obtained by SENSE were used to derive the sorting order here.
14.4.2.3 PECS Applied to CE-MRA
The PECS method has been extended to enable it to be applied to CE-MRA. There is a strong desire in CE-MRA to increase the temporal resolution, that is to increase the acceleration factor in image acquisition and reconstruction. The algorithm is depicted in Fig. 14.9. Note that the depiction is for 3D imaging, so the sampling patterns depicted correspond to the two phase encoding directions. In the first acquisition, a small number of k-space samples (corresponding to a high acceleration factor) are acquired at pseudo-random locations; a second acquisition takes the same number of samples, again randomly distributed, but at a different subset of k-space locations; the acquisitions proceed in this manner until