Shot Rejection for Robust Diffusion Weighted Imaging

The set of equations used for multishot DWI reconstruction were written by Miller et al.¹ (2003) and Liu et al. (2006) and have not been modified for 20-ish years².

Our article describes a modification to the multishot DWI reconstruction that accounts for scenarios where the phase navigator is inaccurate, a problem that occurs in almost every abdominal DWI scan, and sometimes in other anatomies. Several mathematical justifications for this modification are analyzed. We also explicitly illustrate the importance of the phase navigator resolution³, a critical parameter in both acquisition and reconstruction that has been neglected in the literature. I hope that this work will make "robustness to motion" a factor when evaluating new DWI acquisition methods.

Python code that implements linear reconstructions for rejecting shots overly corrupted by motion in diffusion weighted imaging is available here.

Lee PK, Zhou X, Hargreaves BA. “Robust multishot diffusion-weighted imaging of the abdomen with region-based shot rejection,” Magnetic Resonance in Medicine. 2024. doi.org/10.1002/mrm.30102

¹ According to Google Scholar, this article currently has 248 citations (March 2025). There are about 20,000 3T MRI scanners worldwide, and let's assume only 25% do DWI clinically, with 20 patients scanned per day, and only 50% of the protocols include DWI. We can then estimate that the multishot DWI equations are used on ~50,000 patients every day (ignoring 1.5T scanners)! If we consider citations to be a measure of impact, then 248 seems relatively low compared to MRI compressed sensing (~9,000 citations), which is not used nearly as widely in the clinic.

² Multiplexed Sensitivity Encoding (MUSE) (2016) did not change the underlying linear system, but is frequently attributed with having provided the insight that parallel imaging could be used to reconstruct phase navigators. I recently (May 2025) discovered this article by Skare et al. (2007) which applied coil sensitivities estimated from b = 0 s / mm² EPI images to reconstruct each shot independently, followed by magnitude averaging.

³ I also discovered (May 2025) this article by Uecker et al. (2009) that explicitly talks about violation of low-resolution phase assumptions, and has some nice examples in brain (sorry for not citing this). However, they still only used phase navigators with magnitude 1. Interestingly, it also used coil sensitivities estimated from b = 0 s / mm² to reconstruct phase navigators and solve the full multi-shot problem! Perhaps the reason not much attention was paid to this is because EPI was not used, but the contributions from the 2016 MUSE article are essentially all here.