Abstract ID: 115 A Monte Carlo-based eigenspectrum decomposition technique for computed tomography

Abstract

Purpose The characterization of computed tomography (CT) X-ray spectra is important for beam-hardening correction techniques and raw-data reconstruction methods [1] , [2] . In this work, we propose a novel spectrum estimation approach based on transmission measurements and the use of Monte-Carlo (MC) to generate basis spectra. Methods The EGSnrc/BEAM MC code is used to generate basis X-ray spectra. The XTUBE component module is used to produce bremsstrahlung photon energy spectra from monoenergetic electrons transported through a tungsten target (10°), a beryllium window (1 mm) and additional filtration. To ensure that the modelled spectra cover the features of the unknown spectrum, a total of 40 spectra with various levels of filtration are generated using different thicknesses of aluminum and carbon, ranging from 2 mm to 18 mm. Principal component analysis is performed on the MC-generated model spectra to extract a set of linearly independent basis functions, each called eigenspectrum, which reduces the dimensionality of the problem and allows stable fitting. Transmission measurements of a calibration phantom are simulated using ray-tracing with an 80-kV source and added Poisson noise. The estimated spectrum is expressed as the weighted sum of eigenspectra and reconstructed through a constrained least squares technique. Results Using 8 eigenspectra, the 80-kV spectrum is reconstructed with a RMS error (RMSE) of 3.8%. The difference between the mean energy of the estimated and true spectrum is 0.01 keV. Reproducing the same methodology for a 140-kV spectrum yields a RMSE of 4.5% and mean energy difference of −0.1 keV. Conclusion The proposed method is shown promising to accurately characterize X-ray spectra with transmission data. With limited details on the X-ray tube and relying solely on calibration scans, our methodology provides robust spectrum estimation and is promising for reducing the known ill-posedness [3] , [4] of known transmission-based approaches. Applications of the technique to CT are expected to improve the accuracy of quantitative imaging for radiotherapy.