Interior and sparse-view image reconstruction using a mixed region and voxel-based ml-em algorithm

Jingyan Xu, Benjamin M.W. Tsui

Research output: Contribution to journalArticlepeer-review

2 Scopus citations


We propose a new interior region-of-interest (ROI) image reconstruction method for emission tomography. The additional information to make the interior problem uniquely solvable is that a specific region inside the interior ROI is known to have uniform intensity level, but the constant level is unknown. The uniqueness of solution in this situation is analyzed by combining two existing approaches, namely (1) when the full knowledge of a small region inside the interior ROI is known, and (2) when the complete interior ROI is piecewise constant. The image reconstruction is provided by a mixed region and voxel based Poisson likelihood ML-EM algorithm that takes care of the photon statistics and the uniform attenuation effect in emission tomography. This algorithm reconstructs the unknown constant (region-model) and the rest of the interior ROI (voxel-model) simultaneously. The uniqueness result assumes that all line integrals through the interior ROI are acquired. When only a finite number of projection views are available, the mixed region and voxel based ML-EM algorithm can also reduce image artifacts from sparse-view and interior data acquisition in stationary multipinhole SPECT.

Original languageEnglish (US)
Article number6248719
Pages (from-to)1997-2007
Number of pages11
JournalIEEE Transactions on Nuclear Science
Issue number5 PART 1
StatePublished - 2012


  • Exponential radon transform
  • Interior problem
  • ML-EM
  • Molecular imaging
  • Multimodality
  • Multipinhole collimator
  • Poisson likelihood
  • ROI image reconstruction
  • Sparse-view reconstruction
  • Stationary SPECT

ASJC Scopus subject areas

  • Nuclear and High Energy Physics
  • Nuclear Energy and Engineering
  • Electrical and Electronic Engineering


Dive into the research topics of 'Interior and sparse-view image reconstruction using a mixed region and voxel-based ml-em algorithm'. Together they form a unique fingerprint.

Cite this