Single-shot echo planar imaging (ssh-EPI), which is widely used in diffusion-weighted imaging (DWI), has the advantages of short acquisition time and less sensitivity to motion. However, it suffers from geometric distortion, image blurring and signal loss due to local B0 field inhomogeneity, for which imaging resolution is highly restricted, even when parallel imaging is applied, such as sensitivity encoding (SENSE) ^[1]. Zoomed field-of-view (FOV) imaging provides a feasible way to increase in-plane resolution by reducing the FOV along the phase-encoding (PE) direction, which has been approved in previous studies ^{[2, 3, 4]}. To further increase resolution and reduce distortion, it is intuitive to apply parallel imaging to zoomed FOV DWI but at the cost of more SNR reduction and reconstruction artifact. SENSE usually cannot provide satisfactory results in this situation due to the ill-conditioned inverse of SENSE encoding matrix. Hence, the goal of this work is to provide a new reconstruction method for zoomed FOV DWI with parallel acquisitions.

Theory: SENSE reconstruction is formulated as I = S*p, where I is the vector of pixels in each aliased coil image at the same location, S is the matrix of coil sensitivity values, and p is the vector of pixels to be reconstructed. In zoomed FOV imaging, the locations of fold-over pixels get closer and the sensitivity values in these locations become similar, resulting in more amplified noise and artifact in the reconstruction than in the full FOV case due to more ill-conditioned inverse of S. To enhance SNR, more number of signal averages (NSA) are performed traditionally, but they are not fully used in zoomed FOV imaging via just an average. In practice, the image phase varies across different NSAs and the coil sensitivities can be modulated by different phases. In the proposed reconstruction, Joint SENSE across NSA, instead of average after individual SENSE, is performed and the image phase P is extracted after some proper de-noising (total-variation de-noising, for example). Then the new reconstruction can be reformulated as I’ = S’*p, where I’ = [I₁; I₂; …; I_NSA], S’ = [S₁.*P₁; S₂.*P₂; …; S_NSA.*P_NSA]. In this way, the condition of the inverse of S’ is greatly improved compared to S.

Data acquisition: The DWI experiment was performed on a Philips 1.5T MRI scanner (Philips Healthcare, Suzhou, China) with an 8-channel head coil. All the scans in this experiment involved 3 diffusion encoding directions with b = 1000 s/mm² and the PE direction was set to left-right. The ssh-EPI full FOV DWI was first performed using the following parameters: SENSE acceleration factor = 2, FOV = 230*230 mm², default in-plane resolution = 2.3*2.3 mm², slice thickness = 5mm, TR = 2500ms, TE = minimum and NSA = 2. Then, iZOOM^[4] DWI was scanned twice with SENSE acceleration factor set to 1 and 2 respectively, and the other parameters were kept the same in these two scans: in-plane resolution = 1.5*1.5 mm², FOV = 230*60 mm², slice thickness = 5mm, TR = 2500ms, TE = minimum and NSA = 12.

Data processing: Since different SENSE acceleration factors generate different distortion patterns in the images, we artificially under-sampled the fully sampled iZOOM imaging data by a factor of 2. Images were reconstructed using conventional SENSE and the proposed method, and the root-mean-square-errors (RMSE) were calculated for quantitative comparison. For visual comparison, conventional SENSE was performed on all the imaging data in this experiment and the proposed method was conducted on the iZOOM imaging data with SENSE acceleration factor of 2.

Fig. 1 shows the reconstruction results of the simulated imaging data. The conventional SENSE magnitude average introduces strong noise and artifact in the image, whereas higher reconstruction fidelity is achieved using the proposed method. Fig. 2 lists the RMSEs of SENSE magnitude average and the proposed reconstruction in all the 3 diffusion directions, where the reference image is obtained using CLEAR (Philips Healthcare, Best, The Netherlands) with the fully acquired iZOOM imaging data. The RMSE of the proposed method is greatly reduced compared to that of conventional SENSE average. In visual comparison, although in-plane resolution is set to 2.3mm and parallel acquisition is applied, distortion in the anterior fossa (white arrow) is still severe in the full FOV DWI (Fig. 3A). iZOOM DWI with fully sampled data (Fig.3B) is able to provide higher image resolution, but the distortion is still obvious. Fig. 3C and Fig. 3D show the two reconstructions of iZOOM DWI with parallel acquisitions. We can see that the SENSE magnitude average result is covered by strong noise and artifact, while the anatomical structure in the anterior fossa is better revealed in the proposed reconstruction.In this work, we present a new reconstruction method for high-resolution zoomed FOV DWI with parallel acquisitions. The sensitivity values can be modulated by phases from different NSAs so that data from all NSAs can be used to jointly calculate the pixels to be reconstructed. The improved condition of the inverse of S’ makes the reconstruction results with less amplified noise and artifact compared to SENSE. Both quantitative and qualitative results show that the proposed method can provide better reconstruction than conventional SENSE magnitude average.