4D computed tomography reconstruction from few-projection data via temporal non-local regularization

Med Image Comput Comput Assist Interv. 2010;13(Pt 1):143-50. doi: 10.1007/978-3-642-15705-9_18.


4D computed tomography (4D-CT) is an important modality in medical imaging due to its ability to resolve patient anatomy motion in each respiratory phase. Conventionally 4D-CT is accomplished by performing the reconstruction for each phase independently as in a CT reconstruction problem. We propose a new 4D-CT reconstruction algorithm that explicitly takes into account the temporal regularization in a non-local fashion. By imposing a regularization of a temporal non-local means (TNLM) form, 4D-CT images at all phases can be reconstructed simultaneously based on extremely under-sampled x-ray projections. Our algorithm is validated in one digital NCAT thorax phantom and two real patient cases. It is found that our TNLM algorithm is capable of reconstructing the 4D-CT images with great accuracy. The experiments also show that our approach outperforms standard 4D-CT reconstruction methods with spatial regularization of total variation or tight frames.

MeSH terms

  • Algorithms*
  • Humans
  • Imaging, Three-Dimensional / methods*
  • Pattern Recognition, Automated / methods*
  • Phantoms, Imaging
  • Radiographic Image Enhancement / methods
  • Radiographic Image Interpretation, Computer-Assisted / methods*
  • Radiography, Thoracic / instrumentation
  • Radiography, Thoracic / methods*
  • Reproducibility of Results
  • Respiratory-Gated Imaging Techniques / methods*
  • Sensitivity and Specificity
  • Tomography, X-Ray Computed / instrumentation
  • Tomography, X-Ray Computed / methods*