The direct generation of images in a divergent coordinate system may be important to some applications of MR imaging in MRI-radiotherapy hybrids. Specifically, in the circumstance where 2D images are being used for real-time therapy beam guidance, image generation in a divergent coordinate system that matches that of the treatment beam (which originates from a single source point) will reduce targeting errors and inadvertent dosing of critical structures. This work presents a theoretical hardware solution to a variety of MRI-radiotherapy implementations which involve the replacement or augmentation of conventional linear gradients. An experimental and simulated verification is presented.
The encoding gradients must clearly be altered in order to accommodate this new geometry. If the radiation source is fixed relative to the MR unit, only the two gradients perpendicular to the beam direction need be modified, as a divergent geometry can be generated with parallel slice boundaries as seen in Figure 1B. The in-plane gradients will have to modified to provide an increasing signal compression as one moves towards the beam source. This compression will correspond to an increasing in-plane gradient strength, allowing each "divergent" pixel to be encoded with the same gradient field (Figure 2). As opposed to the conventional in-plane gradients which can be described as
$$$G_{i}\left ( x,y,z \right ) \propto \hat{r}_{i} \cdot \left \langle x,y,z \right \rangle$$$,
the increasing gradient strength in the direction of the beam source can be described as
$$$G_{i}\left ( x,y,z \right )\propto \frac{SID}{SID+\hat{r}_{s}\cdot \left \langle x,y,z \right \rangle} \hat{r}_{i}\cdot \left \langle x,y,z \right \rangle$$$,
where $$$\hat{r}_{i}$$$ is a unit vector pointing in the direction of one of the in-plane gradients, and $$$\hat{r}_{s}$$$ is a unit vector pointing from the beam source toward isocentre. $$$SID$$$ represents the distance between the source and isocentre.
This solution will only function when the in-plane gradients are fixed relative to the beam-source. However, in some implementations of these MRI-radiotherapy hybrids, the source is made to rotate about a fixed MRI. In this circumstance there is no one gradient coil that can produce an in-plane BEV gradient for all source orientations. However, despite the fact that the ideal BEV gradient field does not vary linearly with distance to the source, a combination of a conventional linear gradient and a second-order field pattern can approximate the distribution well over small regions of space (Figure 3). The required second-order field pattern will be a product of the distance from isocentre in the direction of the in-plane gradient, and the distance from isocentre in the direction of the source, and its weighting determined numerically based on the slice thickness and position. Two such gradients would be required for any one orientation, with a combination of four providing a basis set for all source positions.
A series of 12-cm gelatin rods were arranged to converge upon an imaginary source 100-cm away as depicted in Figure 4. This phantom was then placed with the base of the untilted rod at isocentre and rotated about this point by an arbitrary 26 degrees. The phantom was imaged with a 12-cm slice using a centre-out radial sequence with a 150-μs TE, and 102 spokes. The sequence was then repeated 102 times with the 2nd-order shim gradients manually configured to create the ideal second-order field for one of the particular spokes. For this orientation the ZY, XZ, 2XY, and X2-Y2 gradients were used as the basis set, and at any one time the combination of shims created a second-order field of 5.263 mT/m2, accompanying the read-encoding gradient with strength of 4.639 mT/m. This manual process was required as shim gradients are not made capable of switching rapidly in conjunction with the linear encoding gradients.
A simulation of the same phantom was acquired using the same radial sequence and reconstructed the same way (filtered back-projection).