Item

Abstract

Introduction In fast imaging techniques, like single-shot TSE sequences [1], the short T2 values of different tissues of interest are comparable to the echo-train duration, which results in significant T2 decay during the k-space acquisition. This leads to a voxel-T2-dependent k-space filter, which results in a complex relaxation-dependent blurring or ringing. We propose an end-to-end optimization approach to reduce T2 blurring in TSE sequences by performing a joint optimization of Flip Angle (FA) design and deblurring CNN. Methods Signal simulation and optimization were performed using a Phase Distribution Graph algorithm [2] implemented in the MRzero framework [3]. For maximal blurring, a single-shot 2D TSE sequence (matrix:128x128, FOV:200mm, GRAPPA2, TEeff = 11.8ms) was used with centric reordering. The forward simulation outputs TSE images (blurry: input to NN) and the transversal magnetization (sharp: target for NN), generated from synthetic brain samples. MRzero optimizes sequence parameters (refocusing FA magnitudes/phases) and DenseNet [4] parameters simultaneously. The final sequence was exported with Pulseq [5] for in vivo measurements of a healthy subject at a Siemens Prisma 3T scanner using a 20ch head coil, comparing to a 16-shot TSE sequence (matrix:128x128, FOV:200mm, GRAPPA2, TR = 12s, TEeff = 11.8ms). Results & Discussion The optimized FA deviate strongly from 180° refocusing FA train with a varying phase pattern (Fig. 2a). In vivo measurements of the final sequence with and without DenseNet processing are shown in Figure 2b and compared to a single-shot and 16- shot TSE sequence with 180° flip angle train, respectively. The T2 blurring in the phase encoding direction (anterior- >posterior) is strongly reduced by optimized FA design and NN. Compared to a 16-shot TSE sequence with 3:12min acquisition time the optimized single shot sequence provides a similar contrast in below 1s. Still, some mismatches e.g. in the thalamus region are visible. Differences in FA trains compared to previous work [6] hints to a new optimal sequence design, when sequence and NN reco are optimized simultaneously.