Phillip Matthew Young MD, Presenter: Nothing to Disclose

Peng Lai PhD, Abstract Co-Author: Employee, General Electric Company

Michael Lustig PhD, Abstract Co-Author: Nothing to Disclose

Marcus T. Alley PhD, Abstract Co-Author: Nothing to Disclose

John Pauly, Abstract Co-Author: Research support, General Electric Company

Garry Evan Gold MD, MSEE, Abstract Co-Author: Consultant, Zimmer Holdings, Inc Consultant, ICON plc Consultant, Cartilix, Inc Research support, General Electric Company Research support, GlaxoSmithKline plc

John De Wolfe MacKenzie MD, Abstract Co-Author: Nothing to Disclose

Shreyas Shreenivas Vasanawala MD, PhD, Abstract Co-Author: Research grant, General Electric Company

00030490-DMT et al, Abstract Co-Author: Nothing to Disclose

We aim to develop and test feasibility of compressed sensing MRI for fast proton density weighted 3D MRI.

A 3D flip angle modulated FSE sequence was modified to sample k-space pseudorandomly, replacing coherent aliasing artifacts with noise-like artifacts. Images could then be reconstructed with a traditional parallel imaging algorithm (ARC) or a compressed sensing non-linear image reconstruction algorithm (CS) that suppressed these artifacts and extracted the true sparse signal components via an L1-norm penalty in the wavelet domain (Lustig M, et al. Sparse MRI: The application of compressed sensing for rapid MR imaging. Magn Reson Med 2007;58(6):1182-95).   With IRB approval and informed consent, 6 patients referred for knee MRI (ages 5y-20y) were scanned with the modified sequence using an 8-channel coil on a GE Signa1.5T scanner. Datasets were acquired with 3 different acceleration factors: 2x2, 2.5x2.5, and 2x3. Each dataset was reconstructed with ARC alone and separately with ARC and the CS algorithm. Two board-certified radiolgists blindly rated images both overall and for 12 anatomic structures in the knee. The readers compared 2x2ARC with 2x2ARC+CS, 2x3ARC with 2.3ARC+CS, 2x3ARC with 2.5x2.5ARC (neither with CS), and 2x3ARC with 2.5x2.5ARC+CS. The comparison recorded whether a structure was more clearly seen with one sequence (+2), a structure was more aesthetically pleasing but not better seen (+1), or was equivalent (0). Differences were tested with a paired Wilcoxon test.

Ratings from the first reader were higher than for the second reader (p<.0001). Agreement between readers was low (kappa = 0.14, p<.0001, 95% CI: 0.06-0.23).  Despite this, 2x2 ARC+CS scored higher than 2x2 ARC (p<.012), and 2x3ARC+CS scored significantly higher than the 2x3ARC (p<.002). While readers showed no preference for 2.5x2.5ARC versus 2x3ARC (neither with CS) (p<.083), there was significant preference for 2.5x2.5ARC+CS versus 2x3 ARC alone (p<.003).   

Readers consistently and significantly preferred images generated with the combined ARC and CS algorithm to those generated with ARC alone. While improvement was often aesthetic, ARC and CS sometimes led to better definition of anatomy than ARC alone.

Isotropic high-resolution volumetric proton density sequences are lengthy. Compressed sensing may permit faster imaging by improving image quality at high acceleration.

Young, P, Lai, P, Lustig, M, Alley, M, Pauly, J, Gold, G, MacKenzie, J, Vasanawala, S, et al, 0, Proton Density-weighted Compressed Sensing MRI.  Radiological Society of North America 2009 Scientific Assembly and Annual Meeting, November 29 - December 4, 2009 ,Chicago IL. http://archive.rsna.org/2009/8013378.html