Recently, Peled and Yeshurun (1) investigated the possibility of calculating superresolution images from a set of spatially shifted, low-resolution images. Eight low-resolution images were acquired with an in-plane shift in phase encoding (PE) direction of 0, 1/4, 1/2, 3/4 pixel, and the same PE-shifts with an additional shift of 1/2 pixel in the read direction.All images were acquired with the same field-of-view (FOV) and resolution. This means that the sampled positions in k-space are identical for all images, since the sampling positions in k-space are uniquely defined by the given FOV and resolution, and vice versa. Let us now consider an in-plane shift along the PE direction. As given by the Fourier-shift theorem a spatial shift in image space is equivalent to a linear phase modulation in k-space along the shift direction (2,3). The sampled positions in k-space do not depend on any in-plane shift along PE direction. As a result, the switched gradient patterns of all three axes as well as the applied RF-pulses are completely identical for any PE-shift. From this it follows immediately that the acquired k-space data is also totally identical, except for the noise and does not depend on the in-plane shift. The actual shift of the image is always done in a postprocessing step, either by multiplying the k-space data K(k m ,k n ) with⌬y is the shift along PE direction, or by appropriate subpixel scrolling of the Fourier transformed k-space data (image). Multiplying or scrolling does not add any new information and is in no way correlated with the imaged object. In principle, the same holds for an in-plane shift along the read direction. Again, all gradient and RF pulses are identical for any shift and thus the positions of the acquired points in k-space are always the same. The shift along read is achieved by a slightly off-resonant receiver frequency, which is again equivalent to a linear phase modulation along the read direction. Thus, no new information can be acquired using different in-plane shifts since an in-plane shift is a postprocessing step and does not influence the measurement procedure (gradients and RF pulses) itself.Of course, images that are shifted by a fraction of one pixel dimension against each other look different and the subtraction of these images is not zero (or pure noise). The images are simply sampled at different positions in image space but the source data is identical.It is, therefore, not possible to increase resolution by acquiring images that are spatially shifted against each other, but it is obviously possible to increase the SNR. The images presented in Ref. 1 clearly show an increase in resolution. However, I think that this increase was probably achieved by the increased SNR and by correcting (amplifying) the T * 2 -decreased signal amplitude at the outer parts of k-space. I presume that the same results can be obtained by acquiring eight images with identical in-plane positions.
REFERENCES1. Peled S, Yeshurun Y. Superresolution in MRI: application to human white matter fibe...