High resolution and high quality diffusion weighted imaging (DWI) can be achieved by using multi-shot EPI techniques. However, DWI data acquired with multi-shot EPI sequence are still susceptible to 1) geometric distortion in the presence of magnetic field inhomogeneities and 2) aliasing artifacts induced by shot-to-shot phase variations. To simultaneously remove geometric distortion and aliasing artifacts without relying on additional phase mapping scan, in this study we develop new segmented DW-EPI sequence and reconstruction framework. Through reversing the polarity of the phase-encoding gradient between interleaves in DW-EPI sequence, field inhomogeneities and shot-to-shot phase variations can be inherently estimated using SENSE reconstruction without additional field mapping scan. With these estimated information, DWI images free from distortions and aliasing artifacts can be reconstructed using multiplexed sensitivity encoding MRI framework based on projection onto convex sets and multiplexed sensitivity encoded MRI (POCSMUSE) ^1,2.

Data acquisition: Simulated DWI data were acquired with a modified 4-shot interleaved spin-echo EPI sequence, in which the polarity of the phase-encoding gradient of the 1st and 3rd shots was ‘bottom-up’, and that of the 2nd and 4th shots was ‘top-down’. This reversed and interleaved acquisition scheme provided two geometric distortion patterns from which field inhomogeneities could be directly estimated ³.

A hybrid 4-shot DW-EPI simulation was performed to evaluate the new reconstruction framework: First, field inhomogeneities (-50 to 100 Hz) were included. Second, four predefined phase maps were mathematically added to he the scanned object, reflecting motion-induced shot-to-shot phase variations in DWI scans. Third, a set of simulated EPI based DWI data (with motion induced shot-to-shot phase variations) was produced using our reversed-gradient 4-shot interleaved EPI sequence, for all of the 8-channel coil elements. Scan parameters included: TE/TR=60/4000ms, echo-spacing time (ESP)=0.62ms, FOV=24 cm x 24 cm, matrix size = 256 x 256.

Data reconstruction: First, images free from motion-related artifacts (but geometrically distorted) were reconstructed from each segment using the conventional SENSE algorithm ⁴. Second, segmented-specific phase variations were calculated from images produced in step 1 and then spatially smoothed ¹. Third, field inhomogeneities were calculated from distorted images produced in step 1. Fourth, we used a newly developed POCSMUSE framework ² to reconstruct DWI data with the following steps, as shown in Figure 1: 1) starting with an initial guess of undistorted and un-aliased source image (P⁰); 2) applying the sensitivity profiles; 3) applying phase modulation ⁵ to produce a set of distorted images based on field inhomogeneities; 4) applying segment-specific phase variations to produce a set of simulated images (D_i,j); 5) replacing parts of the simulated data with experimentally acquired k-space data (i.e., data projection); 6) demodulating the projected data with sensitivity profiles and known shot-to-shot phase variations; 7) applying inverse phase modulation to generate an updated source image Pn, which was further used as the input of step 1 until the iterative processes converge. Afterward, an image with reduced geometric distortion and motion-induced aliasing artifacts was generated.

Figure 2 shows DWI images obtained 2D Fourier transform reconstruction of all the k-space data (left panel), MUSE reconstruction of the k-space data of the 2nd and 4th segments (i.e. data acquired with ‘top-down' gradient: middle panel), and POCSMUSE reconstruction of full k-space data (right panel). As shown in the left panel, the uncorrected image is significantly affected by artifacts. In the middle panel, the motion-induced aliasing artifacts are removed by MUSE, while the reconstructed image is still distorted. As shown in the right panel, the POCSMUSE framework can simultaneously and effectively correct both geometric distortions and aliasing artifacts. The developed framework has two strengths: 1) it effectively corrects geometric distortions and phase variations without relying on extra scan; 2) the multiplexed sensitivity based reconstruction ^1,2 is much less susceptible to undesirable noise amplification than conventional parallel imaging reconstruction. In conclusion, the integration of a novel interleaved DW-EPI pulse sequence and the POCSMUSE algorithm can produce high-quality DWI data free from geometric distortions and aliasing artifact due to shot-to-shot phase variation.