Respiratory motion remains a major challenge to the use of MRI in the body. This is most notable in dynamic contrast enhanced (DCE) imaging of the chest and abdomen, which requires continuous imaging to capture bolus dynamics. While numerous motion-insensitive, free-breathing, accelerated 3D imaging strategies have been proposed^1-2, they remain sensitive to complex respiratory patterns. The alternative use of rapid 2D multi-slice imaging would be appealing due to its ability to freeze motion; however, 2D multi-slice imaging is highly sensitive to in-flow and steady state preparation. Recently, sliding 2D slices were proposed as a means to alleviate these artifacts and capture images in which motion artifacts manifest as correctable geometric distortions³. A drawback to this technique is the need for highly accelerated 2D spiral imaging which is sensitive to off-resonance artifacts. In this work, we investigate the synergistic combination of simultaneous multi-slice, sliding slices, pseudo random Cartesian sampling with post-processing registration to allow for high frame rates, lower demands on in plane acceleration, and FOV tailoring.

A pulse sequence diagram is shown in Figure 1. As in the initial spiral based implementation, imaging was performed by interleaving a 3D saturation pulse with a 2D imaging module. Data is sampled as shown in Fig 2 with data collected during a continuous sweep of the excitation frequency of a multi-band RF pulse. During this sweep, readouts are collected with ky phase encodings utilizing pseudorandom sampling, based on the golden ratio⁴. As viewed as a function of time in Figure 2b, this creates a set of images with overlapping replicates of the underlying image. The signal of each sample is modeled as a function of the k-space position (kx,ky) and the location of the slices (z1,z2):

$$d(k_x,k_y,z_1,z_2)=\int_{}^{}\int_{}^{}\int_{}^{} S(x,y,z)e^{i2\pi(xk_x+yk_y)}\left[G(z-z_1)I_{t+1}+G(z-z_2)I_{t}\right]$$

where G is the slice excitation profile; S is the sensitivity map; and It is the image of time frame t. Making a discrete approximation of Eq1, images can be recovered utilizing a penalized least squares, with the penalty being a smoothness constraint to prevent super resolution effects (i.e. the resolution in z should not exceed the slice thickness). Note due to the interleaved nature of the data in time and space, the reconstruction of any given slice requires the reconstruction of all slices at all time points. To evaluate feasibility, phantom and healthy volunteers (after IRB approval and informed consent) were imaged on a clinical 3.0T system (MR750, GE Healthcare, WI, USA). Images were first collected from the ADNI phantom placed in an 8-channel head coil. Sagittal images were acquired with 20mm, ~8Hz sinusoidal S/I motion of the table with 2D, 3D, and simultaneous sliding slice. Sliding slice parameters included: resolution=1.5x1.9x4mm3,TE/TE=1.3/3.8ms, flipimaging=15°, flipsat=8°, with 2 (ky) x 2 (slice) parallel imaging acceleration, 8s temporal footprint, and 20 time frames. 2D and 3D sequences were matched to the same temporal footprint and spatial resolution. Apparent CNR was calculated using the ratio of the highest CNR sphere to that of the background. Dynamic in-vivo free breathing, DCE imaging of the entire thoracic cavity was subsequently performed during bolus passage of a 0.1mmol/kg dose of gadobenate dimeglumine with a 32-channel torso array coil (Neocoil, Pewaukee, WI, USA). Imaging parameters were identical to those used for phantom experiments. Figure 3a shows images acquired without slice unwrapping, with the temporal dimension along the horizontal axis. Periodic motion is clearly visible with two replicates, representing the two slices. Replicates were successfully unwrapped, creating images at twice the frame rate. Figure 3b shows 2D, 3D and sliding images. Motion causes significant artifact and ghosting in the 3D images. 2D imaging effectively freezes motion; however, images suffer from lack of contrast and stair stepping artifacts. The sliding approach maintains high contrast and motion insensitivity, which is reflected in high apparent CNR (23.2) compared to 2D (1.15) and 3D (8.34). Figure 4, shows the source sagittal images of in-vivo feasibility testing. Images demonstrate complete freezing of motion, including Cardiac contraction. Reformatting the images into the coronal plane, as in Figure 5b, shows the manifestation of motion in this case with the motion state being different in the R/L dimension. This type of motion is potentially correctable via non-rigid registration, preliminarily shown in Figure 5b. The combination of simultaneous multi-slice imaging with sliding slices allows additional acceleration that translates to higher temporal frame rates. The incorporation with Cartesian sampling facilitates FOV tailoring and increases robustness to off-resonance artifacts. This allows for sagittal acquisitions, casting most respiratory motion in-plane, which is amenable to correction via registration.