Introduction

Clinical diffusion-weighted imaging (DWI) is commonly based on echo-planar imaging (EPI) and suffers from susceptibility- and diffusion-gradient-direction-oriented eddy-current-induced image distortions, which often results in nondiagnostic image quality in local high susceptibility areas of the brain and potentially lead to misdiagnosis (e.g., a false negative or false positive result). Recent technical advances such as parallel imaging¹, reduced FOV DWI², and multi-shot approaches^3,4 are improvements over routine EPI-based DWI, however, the degree of distortion remains prominent at 3T, especially in proximity to air-tissue interfaces. A novel distortion-free imaging scheme, termed DIADEM (Distortion-free Imaging: A Double Encoding Method) was recently introduced and optimized on a high-performance Compact 3T scanner^5,6 and showed robust diffusion imaging in susceptibility prone regions. In this study, we implemented DIADEM-DWI on two clinical 3T scanners and further characterized its performance relative to conventional EPI-based DWI of the brain.

Methods

Implementation of online DIADEM reconstruction: The flowchart of online DIADEM reconstruction pipeline is presented in Fig. 1. Due to its new imaging strategy that is not supported by vendor reconstruction, the automatic DIADEM reconstruction pipeline has been developed and is performed under MATLAB compiler runtime (MCR) environment on a secure external high-performance reconstruction server located outside the MRI network segment utilized with 16 multicore computing processors. All network communications are initiated scanner-side and utilize outbound communication pathways only to avoid any security issue – i.e., the server is incapable of initiating a connection to the scanner. Sequential reconstruction is conducted on the server to reconstruct all the DIADEM scans from multiple scanners. As provided in vendor EPI-DWI reconstruction, a trace (i.e., combined DW images) and apparent-diffusion-coefficient (ADC) map are also calculated in the DIADEM-DWI reconstruction.
Feasibility study for routine clinical practice: DIADEM-DWI was first evaluated for clinical use on a commercially available (60 cm) whole-body 3T scanner with a gradient performance of 50 mT/m and 200 T/m/s (MR750, GE Healthcare) and a standard 8-channel coil (Invivo Corp). Under IRB-approved protocols, first ten patients scheduled for a clinical brain MRI exam with standard diffusion (i.e., EPI-DWI) were recruited to additionally obtain the DIADEM-DWI scan for comparison. Next, the comparison study was translated to a wide-bore (70 cm) 3T scanner with a lower gradient performance of 33 mT/m and 120 T/m/s (MR750W, GE Healthcare) using a 32-channel head coil (Nova Medical). Clinical EPI-DWI scan was obtained on both whole-body 3T scanners with following imaging protocols: FOV=220×220 mm², matrix size (readout×phase-encoding)=128×256, image resolution=1.72×0.86, 38 slices, slice thickness=4 mm, one non-DW (i.e., T2) and three DW images with b=1000 mm²/s, and scan time=1:00~1:10. The identical protocols were applied in DIADEM-DWI scans on MR750/MR750W 3T scanners, except for the matrix size=220×120/200×200, image resolution=1×1.83/1.1×1.1 mm², multiband=1/2, and scan time=4:00/3:59.

Results and Discussion

Strong local distortions are clearly visible in EPI-DWI near boundaries of temporal and frontal lobes on a clinical whole-body 3T scanner (Fig. 2). In distinction, negligible distortion was observed in DIADEM and the images matched very well to the geometrically correct reference (i.e., fast-spin-echo) image. Since this approach is immune to both T2* blurring and eddy-current-induced distortions varying according to the applied (i.e., three) diffusion directions, the images were visually much sharper than EPI even though the intrinsic image resolution in EPI was slightly higher (Fig. 2). In addition, additional magnetic susceptibility artifacts induced by metallic shunt components were effectively resolved in DIADEM imaging (Fig. 3). Compared to the standard EPI-DWI, DIADEM-DWI produced more reliable diffusivity coefficients, even in local brain structures with high susceptibility effects including internal auditory canal (Fig. 4A), cisternal segment of the trigeminal nerve (Fig. 4A), and optic nerve (Fig. 4B). Acquisition time for DIADEM-DWI was around 4 minutes compared to 1:10 minutes for EPI-based DWI, but still acceptable for clinical use. High image resolution on DIADEM was possible using multiband imaging⁷ on the wide-bore 3T scanners, even with the lower gradient performance MR750W scanner (Fig. 5). The reconstruction time including the data transfer time over network were around 7 and 15 minutes, respectively without and with multi-band imaging protocol, which could be further reduced by translating the MATLAB- to C-based reconstruction algorithm in future.
DIADEM-DWI provided improved depiction and conspicuity of brain structures compared to standard EPI-DWI. DIADEM permitted more confident diagnosis of small acute brain infarctions in local high-susceptibility areas that are commonly challenging to interpret, primarily because of fewer artifact-related false positive and false negative results. Visualization of cranial nerves, orbital and suprahyoid neck structures was markedly improved with DIADEM. Evaluation of postoperative DWI changes was also superior with DIADEM-DWI due to reduced image degradation related to post-surgical susceptibility changes and areas of pneumocephalus.

Conclusion

DIADEM high-resolution distortion-free DWI was successfully implemented onto clinical 3T scanners and demonstrated marked improvements in diffusion signal characterization compared to the currently implemented EPI-based DWI, especially in local high-susceptibility regions.

Acknowledgements

This work was supported by NIH U01 EB024450, NHI U01 EB026979, and Mayo Clinic Imaging Biomarker Discovery Program.