Faster algorithm for NLM based image reconstruction

Recently, non-local means (NLM) based image reconstruction methods showed significant improvement in image quality and quantitation in medical imaging such as PET, SPECT, CT, and MR. However, using NLM based regularizer in image reconstruction resulted in too slow convergence. We proposed a novel optimization method based on alternating direction method of multiplier (ADMM) to improve the speed of convergence for NLM based reconstruction. Compared to other previous methods such as gradient descent (GD), expectation-maximization (EM), ordered subset EM (OSEM), and quasi-Newton method (L-BFGS-B), our ADMM method substantially improved the speed of reconstruction or reduced the time to reach the minimum root mean square error (RMSE) as shown in the figure. This work has been published in IEEE Transactions on Medical Imaging (IF=3.799, 5-year IF=4.575). 

  1. S Y Chun, Y K Dewaraja, J A Fessler, “Alternating direction method of multiplier for emission tomography with non-local regularizers,” Proc. Intl. Mtg. on Fully 3D Image Recon. in Rad. and Nuc. Med, pp. 62-5, 2013.
  2. S Y Chun, Y K Dewaraja, J A Fessler, “Alternating direction method of multiplier for tomography with non-local regularizers,” IEEE Transactions on Medical Imaging, 33(10):1960-8, Oct. 2014.
Advertisements

Comments are closed.

Create a free website or blog at WordPress.com.

Up ↑