A theoretical comparison of tissue parameter extraction methods for dual energy computed tomography

• Published in: Medical physics (Lancaster) 2014-08, Vol.41 (8Part1), p.081905-n/a
• Main Authors: Tremblay, Jean‐Étienne, Bedwani, Stéphane, Bouchard, Hugo
• Format: Article
• Language: English
• Subjects: Biological material, e.g. blood, urine; Haemocytometers; biological tissues; Calibration; Computed tomography; Computer Simulation; Computerised tomographs; computerised tomography; Conformal radiation treatment; Databases, Factual; Digital computing or data processing equipment or methods, specially adapted for specific applications; dual energy; effective atomic number; electron density; Electrons; feature extraction; Humans; Image data processing or generation, in general; image reconstruction; Medical image noise; medical image processing; Medical image reconstruction; Models, Theoretical; phantoms; Phantoms, Imaging; Photons; radiation therapy; Reconstruction; stoichiometric calibration; Tissues; Tomography - instrumentation; Tomography - methods; X‐ray spectra
• ISSN: 0094-2405; 2473-4209
• DOI: 10.1118/1.4886055

Purpose: To evaluate the reliability of common sinogram‐based DECT reconstruction methods for radiotherapy tissue characterization and to evaluate the advantage of combining them with a stoichiometric calibration. Methods: The sinogram‐based DECT method defined by Alvarez and Macovski [“Energy‐selective reconstructions in x‐ray computerized tomography,” Phys. Med. Biol. 21, – (1976)] is adapted to the XCOM photon cross sections database and also generalized to a two‐material decomposition method. A theoretical framework is developed using a test phantom containing human tissue compositions for comparing the sinogram‐based methods and the calibration‐based method, being defined as the application of the stoichiometric calibration technique of Bourque et al. [“A stoichiometric calibration method for dual energy computed tomography,” Phys. Med. Biol. 59, 2059–2088 (2014)] on monoenergetic images being generated with a sinogram‐based method. Applying a bias correction to the sinogram‐based method, its performance in extracting human tissue parameters in the presence of noise as well as by altering the photon energy spectrum is compared to the calibration‐based method. Results: In the absence of noise and without spectrum alteration, the calibration‐based method is found to have no benefit on the sinogram‐based method. However, the calibration‐based method is shown to be potentially more reliable than bias‐corrected sinogram‐based methods in situations comparable to the clinical environment, where noise is present and the photon energy spectra can differ from what is used during image reconstruction. In determining electron density, the performance of all methods is comparable in the presence of noise only. Moreover, combined with heavy spectrum alteration, the mean errors on electron density are found higher in sinogram‐based methods in comparison with the calibration‐based method, with 1.2% versus 0.2%. In the presence of significant noise, bias‐corrected sinogram‐based methods yield mean errors on effective atomic number of about 2.5%, as compared to 0.5% for the calibration‐based method. When combined with heavy spectrum alteration, bias‐corrected sinogram‐based methods can lead to error of up to 4% on the effective atomic number versus 1.8% for the calibration‐based method. Conclusions: While sinogram‐based methods have the advantage of eliminating beam hardening effects, results of this study suggest improvements in the accuracy and reliability of extract