B0 inhomogeneity is a major challenge encountered in body diffusion-weighted (DW) echo-planar imaging (EPI), leading to severe signal loss and distortion. The numerous air-tissue interfaces in the torso, the difficulty to accurately shim large volumes and the presence of fat often result in suboptimal image quality. The use of 2D RF pulses in conjunction with parallel imaging (PI) for coronal whole-body DW-EPI has been described¹. While the 2D spatial selectivity can both improve PI performance by phase encoding along the direction with the largest variation in coil sensitivities and reduce distortion by limiting the FOV to the anatomy of interest, the low bandwidth of the 2D RF pulse makes this technique extremely sensitive to off-resonance, resulting in severe signal loss². We present a method that can uniformly excite the whole imaging volume while correcting residual off-resonance-induced distortion in clinically acceptable scan times. Phantom and volunteer experiments were performed to demonstrate feasibility. 2D RF pulses can be used to control the extent of the excited volume along both the slice-select and phase-encoding direction. Here, the blipped direction during excitation corresponds to the slice-select direction during imaging, so that only the main excitation lobe is refocused by the 180° refocusing pulse³. Due to the low slice-select bandwidth of the 2D RF pulse (220Hz) and the relatively high bandwidth of the refocusing pulse (1.2kHz), off-resonance spins are shifted outside of the volume refocused by the 180º, so that the combination of the 2 pulses effectively produces a 2D spatial/1D spectral excitation of finite bandwidth. By progressively shifting the center frequency, the whole range of off-resonance can be excited, as demonstrated in Figure 1, where a 30Hz/cm shim gradient was used to simulate off-resonance. Root-sum-of-squares (RSOS) combination of the spectral components restores signal over the whole FOV but does not correct off-resonance-induced distortion. However, knowledge of the spectral components on a pixel-by-pixel basis allows estimation of the off-resonance field (i.e. field map), which, in turn, can be used to correct off-resonance induced distortion (d=ΔB₀/prBW; d=shift along phase-encode direction in pixels, prBW=pseudo receiver bandwidth per pixel). A center of mass (COM) approach was used to estimate the field map from the frequency “bin” images⁴ (cfr. Figure 1). Phantom experiments were performed to assess accuracy of the estimated field map and effectiveness of the distortion correction algorithm. A separate IDEAL acquisition was performed to obtain a reference field map (6 TEs; TE1/DTE=0.9/0.7ms; flip angle=3º; BW=111kHz; FOV=44cm; phase FOV=50%; 7mm slice thickness; matrix size=160×160; 16 slices). The known shape of the phantoms was used to evaluate the performance of the distortion correction algorithm. Healthy volunteers were scanned using the upper 20 elements of a 32-channel receive-only phased-array coil (NeoCoil, Pewaukee, WI) and an 8-channel receive-only neurovascular coil (GE Healthcare, Waukesha, WI) to demonstrate feasibility. All imaging was performed at 3T (GE MR750, Waukesha, WI).Figure 2 shows RSOS and multispectral (MS) reconstructions using IDEAL and estimated field maps. Both field maps were able to successfully correct the distortion induced by the shim gradient. Note that distortion in the original acquired images is reflected in the COM field map, while the IDEAL-generated field is undistorted. In vivo experiments showed similar off-resonance distributions across different subjects for a specific anatomy. In all cases, 4-5 frequency bins were sufficient to cover the range of off-resonance, which could be achieved with scans under five minutes. Figure 3 shows on-resonance and composite MS images of 2 axial abdominal slices. Note that in this case the frequency spread was minimal in the first slice but large enough to cause significant signal loss, successfully recovered by the MS reconstruction, in a different slice (arrow). Figure 4 and 5 show coronal and axial images of the brachial plexus, where conventional on-resonance imaging was non-diagnostic. Overall, good suppression of on-resonance fat was achieved in the MS images (cfr. RSOS, Figure 4), although some off-resonance fat (arrow) remains due to errors in the estimated field map, that could be corrected by acquiring more frequency bins while keeping scan time constant by reducing the number of signal averages per bin. We developed a DW-EPI method to reduce off-resonance-induced signal loss and distortion, which are especially problematic for body applications. The main limitation consists in the longer duration of the acquisition (proportional to the number of acquired bins). While a few bins are usually sufficient to cover the whole off-resonance range, multi-slice capabilities can be added to the proposed method to extend coverage/reduce scan time.