Diffusion imaging has been widely used in clinical applications. By far single-shot EPI is the most commonly used sequence due to its high speed and insensitivity to bulk motion and physiologic movement. However, this technique suffers from image blurring caused by signal decay during the long acquisition window and static field related problems, e.g., geometric distortion and susceptibility artifact caused by B0 inhomogeneity. Comparing to EPI based methods, diffusion weighted (DW)-PROPELLER sequence, which is based on Turbo Spin Echo (TSE) sequence and PROPELLER(BLADE) trajectory, has the advantage of reduced sensitivity to B0 inhomogeneity and T2 decay induced image blurring [1]. However the application of this sequence is limited by its relatively long scan time, high SAR and violation of CPMG condition. X-PROP [2] with multi-blade k-space filling strategy, was developed to speed up the acquisition and reduce SAR. In this technique the non-CPMG problem is mitigated by phase cycling the refocusing RF pulses, which requires the flip angle of refocusing pulses as close to 180° as possible. Therefore, SAR is still high especially at high fields and B1 field variation may result in unstable image quality. This technique also suffers from off-resonance artifacts when large acceleration factor is applied, due to the contribution from more gradient echoes. The SPLICE (split acquisition of fast spin-echo signals) technique can address the non-CPMG problem reliably by separating two signal components to avoid destructive phase interference[3]. Therefore, we propose a novel technique by sampling two separated echoes with multi-blade (X-PROP) strategy and placing the blades to a single k-space dataset, and demonstrate its application for fast and robust DWI. With this technique, the non-CPMG problem can be effectively mitigated with low-flip-angle refocusing pulses to reduce SAR. Moreover, the off-resonance artifacts can also be reduced with high acceleration factor.

The diagram of the proposed sequence is shown in figure 1. Note that time (t) starts at the end of diffusion preparation (not shown). At this point the spins lie in the transversal plane with spatially variant and unknown phase, which violates the CPMG condition. The signal components with different phases are split into two groups (E1 and E2), as the first spin echo is refocused asymmetrically with regard to the interval between consecutive refocusing pulses. Multi-blade(X-PROP) acquisition strategy is applied to sampling these echoes. EPI-like readout gradients and blips are generated on both readout and phase encoding directions, producing radial-like k-space lines which are placed into separate blades. The readout is repeated throughout the whole echo train, producing all k-space lines for each blade. These blades spans in one single k-space, as shown in figure 2. The echoes from E1 and E2 group have different phases, which are removed by the blade-by-blade phase correction in the following image reconstruction step.

A phase insensitive preparation is also applied to push E1 and E2 to reach steady state as soon as possible by applying a dephasing gradient right after the end of diffusion preparation [4]. Additional rephrasing and dephasing gradients must be added before and after the EPI readout, respectively. Note that the signal acquisition starts from the second interval during which E2 is formed.

This prototype sequence was implemented on a Siemens MAGNETOM Spectra 3T scanner and validated by in vivo experiment. For comparison, conversional DW single-shot EPI and TSE-based T2W images were also acquired.

3 out of 25 slices of the images acquired by single-shot EPI and the proposed method with b = 1000 s/mm² are shown in figure 3. The geometric distortion, susceptibility artifact and blurring were obvious in EPI images while alleviated in the images acquired with proposed method.

Figure 4 shows images acquired at basal brain area where B0 inhomogeneity is severe. Images acquired by the proposed sequence (b = 0 s/mm2) (figure 4b) corresponded well to the TSE images (figure 4a), exhibiting no obvious off-resonance artifacts. This is benefit from more contributions from spin echo signals even high acceleration factor is used.

SPLICE technique is robust for overcoming the non-CPMG problems, while multi-blade technique is highly effective for data acquisition. Our study shows that the proposed sequence can reliably produce diffusion weighted images with fast speed, minimized artifact and moderate SAR. It offers the potential for robust high quality diffusion application.