Compressive computed tomography image reconstruction with denoising message passing algorithms

Alessandro Perelli, Mike E. Davies

Research output: Chapter in Book/Report/Conference proceedingConference contribution

11 Citations (Scopus)


In this paper we address the compressive reconstruction of images from a limited number of projections in order to reduce the X-ray radiation dose in Computed Tomography (CT) while achieving high diagnostic performances. Our objective is to study the feasibility of applying message passing Compressive Sensing (CS) imaging algorithms to CT image reconstruction extending the algorithm from its theoretical domain of i.i.d. random matrices. Exploiting the intuition described in [1] of employing a generic denoiser in a CS reconstruction algorithm, we propose a denoising-based Turbo CS algorithm (D-Turbo) and we extend the application of the de-noising approximate message passing (D-AMP) algorithm to partial Radon Projection data with a Gaussian approximation of the Poisson noise model. The proposed CS message passing approaches have been tested on simulated CT data using the BM3D denoiser [2] yielding an improvement in the reconstruction quality compared to existing direct and iterative methods. The promising results show the effectiveness of the idea to employ a generic denoiser Turbo CS or message passing algorithm for reduced number of views CT reconstruction.
Original languageEnglish
Title of host publication2015 23rd European Signal Processing Conference (EUSIPCO)
Number of pages5
ISBN (Print)978-0-9928-6263-3
Publication statusPublished - 4 Sept 2015
Event2015 23rd European Signal Processing Conference (EUSIPCO) - Nice, France
Duration: 31 Aug 20154 Sept 2015


Conference2015 23rd European Signal Processing Conference (EUSIPCO)


  • Computed tomography
  • Signal processing algorithms
  • Radon
  • Message passing
  • Image reconstruction
  • X-ray imaging
  • Approximation algorithms


Dive into the research topics of 'Compressive computed tomography image reconstruction with denoising message passing algorithms'. Together they form a unique fingerprint.

Cite this