Conventional whole body DWI suffers from spatial misalignment and signal drop off. This hinders accurate fusion between PET/MR images and prevent good visualization of the spine & lymph node, which are critical for interpretation of tumors or metastases. In this study, we proposed a combination of a B0 correction technique & an image registration approach to improve this issue.

Single shot EPI acquisition is commonly used for WB DWI, however, distortion, misalignment and signal loss are often observed due to B0 inhomogeneity [1]. Previous study has described a fast realtime b0 correction method (RTB0) for detecting and adjusting the optimal center frequency(CF) for each slice, resulting in a reduction of the misalignment due to B0 inhomogeneity[2]. However, in area where there is large in-plane variation of B0 field/tissue susceptibility, such as brachial plexus, jaw and skull base, the global CF detected for a slice might be suboptimal. Therefore, we proposed the following techniques (Fig 1):

1. RTB0 with utilization of multi-coil statistics: Optimal CF were computed for each coil element of each slice. Slices with high in-plane CF variation were automatically identified and inter-slice CF fitting avoided these slices to prevent over-correction.

2. Image registration: A two-step registration technique was used to register DWI data with MR Attenuation Correction data (MRAC). MRAC data was based on 3D LAVA-FLEX acquisition, and was spatially comparable to PET data. In this technique, a coarse registration was first used to roughly align the stations, followed by finer step elastic registration applied between the DWI and MRAC.

To validate this approach, WB MRAC and WB DWI with and without RTB0 were performed on 18 consented healthy volunteers (11 Male, 7 Female, Age: 18-80 year old, Height:58-74”, Weight:128-264lb) on a GE 3T 70cm bore PET/MR system using the GEM whole body surface coils. Optimized DWI parameters were: FOV:46cm(LR)x33.6cm(AP), Matrix: 96(freq) x 128(phase), TR/TE:4125ms/65ms, single spin echo, BW:250kHz, Slice thickness:5mm, # slices:50/station (i.e. 25cm per station), STIR, TI=245ms, b-value=0s/mm2(3 NEX), 800s/mm2 (8 NEX), diffusion encoding: 3-in-1, scan time: 2:29min per station. The datasets with RTB0 applied were then processed with the proposed image registration pipeline. Two reviewers (experience technologists) reviewed the images and performed independent rating of the alignment quality based on the rating scale 1-4 (1: significant misalignment, 2: small misalignment, 3: Aligned but unnatural curvature of the spine, 4. Aligned and accurate spine curvature). Image ratings were compared using paired Wilcoxon signed rank test.

Figure 2 shows the alignment improvement of DWI as compared to a saggital localizer when RTB0 is used. A improvement in spatial accuracy can be observed by the acquisition correction technique alone. Figure 3 shows the image alignment rating of 1) Conventional DWI, 2) DWI with RTB0, 3) DWI with RTB0 & Image registration. We observed a progressive improvement with RTB0 technique (2.06±0.68 vs 2.63±0.72, p=0.0008), and further improvement with additional image registration steps (2.63±0.72 vs 3.37±0.60, p=0.00009). Fig 4 shows an example of the complete pipeline. In addition, as the RTB0 technique used a more optimal center frequency for each slice, we also observed a reduction in signal drop off at brachial plexus, and improvement of fat suppression(Fig 5). We also observed no significant changes in ADC values as a result of this approach.The results have demonstrated that spatial accuracy can be improved using a combination of RTB0 & image registration. RTB0 improves the signal uniformity & fat suppression, and therefore also helps improve the registration accuracy. Furthermore, since an additional 3D rigid registration step will be performed on the standard GE PET/MR fusion software, if we eliminate the non-rigid misalignment on MR DWI data, the PET/MR fusion would result in anatomically well-matched fused data. There is also added advantage of improved fat suppression & signal uniformity by the use of RTB0, which is often critical to recover signal at spinal, skull base and lymph node, where metastases commonly occur. In this work, we demonstrated improvement in spatial alignment of WBDWI using a combination of RTB0 & image registration techniques. We validated this technique on volunteers and also demonstrated the benefit of slice-wise center frequency adjustment in fat suppression quality & signal uniformity. This technique would be useful in clinical PET/MR evaluation to ensure spatial alignment between PET & MR data.