Efficient scatter distribution estimation and correction in CBCT using concurrent Monte Carlo fitting G. J. Bootsmaa) Radiation Medicine Program, Princess Margaret Cancer Centre, Toronto, Ontario M5G 2M9, Canada
F. Verhaegen Department of Radiation Oncology (MAASTRO), GROW—School for Oncology and Developmental Biology, Maastricht University Medical Center, Maastricht 6201 BN, The Netherthlands and Medical Physics Unit, Department of Oncology, McGill University, Montreal, Quebec H3G 1A4, Canada
D. A. Jaffray Radiation Medicine Program, Princess Margaret Cancer Centre, Toronto, Ontario M5G 2M9, Canada; Ontario Cancer Institute, Princess Margaret Cancer Centre, Toronto, Ontario M5G 2M9, Canada; and Department of Radiation Oncology, University of Toronto, Toronto, Ontario M5G 2M9, Canada
(Received 29 May 2014; revised 22 August 2014; accepted for publication 29 September 2014; published 16 December 2014) Purpose: X-ray scatter is a significant impediment to image quality improvements in cone-beam CT (CBCT). The authors present and demonstrate a novel scatter correction algorithm using a scatter estimation method that simultaneously combines multiple Monte Carlo (MC) CBCT simulations through the use of a concurrently evaluated fitting function, referred to as concurrent MC fitting (CMCF). Methods: The CMCF method uses concurrently run MC CBCT scatter projection simulations that are a subset of the projection angles used in the projection set, P, to be corrected. The scattered photons reaching the detector in each MC simulation are simultaneously aggregated by an algorithm which computes the scatter detector response, SMC. SMC is fit to a function, SF , and if the fit of SF is within a specified goodness of fit (GOF), the simulations are terminated. The fit, SF , is then used to interpolate the scatter distribution over all pixel locations for every projection angle in the set P. The CMCF algorithm was tested using a frequency limited sum of sines and cosines as the fitting function on both simulated and measured data. The simulated data consisted of an anthropomorphic head and a pelvis phantom created from CT data, simulated with and without the use of a compensator. The measured data were a pelvis scan of a phantom and patient taken on an Elekta Synergy platform. The simulated data were used to evaluate various GOF metrics as well as determine a suitable fitness value. The simulated data were also used to quantitatively evaluate the image quality improvements provided by the CMCF method. A qualitative analysis was performed on the measured data by comparing the CMCF scatter corrected reconstruction to the original uncorrected and corrected by a constant scatter correction reconstruction, as well as a reconstruction created using a set of projections taken with a small cone angle. Results: Pearson’s correlation, r, proved to be a suitable GOF metric with strong correlation with the actual error of the scatter fit, SF . Fitting the scatter distribution to a limited sum of sine and cosine functions using a low-pass filtered fast Fourier transform provided a computationally efficient and accurate fit. The CMCF algorithm reduces the number of photon histories required by over four orders of magnitude. The simulated experiments showed that using a compensator reduced the computational time by a factor between 1.5 and 1.75. The scatter estimates for the simulated and measured data were computed between 35–93 s and 114–122 s, respectively, using 16 Intel Xeon cores (3.0 GHz). The CMCF scatter correction improved the contrast-to-noise ratio by 10%–50% and reduced the reconstruction error to under 3% for the simulated phantoms. Conclusions: The novel CMCF algorithm significantly reduces the computation time required to estimate the scatter distribution by reducing the statistical noise in the MC scatter estimate and limiting the number of projection angles that must be simulated. Using the scatter estimate provided by the CMCF algorithm to correct both simulated and real projection data showed improved reconstruction image quality. C 2015 American Association of Physicists in Medicine. [http://dx.doi.org/10.1118/1.4903260] Key words: x-ray scatter, cone-beam CT, Monte Carlo, image quality
54
Med. Phys. 42 (1), January 2015
0094-2405/2015/42(1)/54/15/$30.00
© 2015 Am. Assoc. Phys. Med.
54
55
Bootsma, Verhaegen, and Jaffray: Efficient scatter distribution estimation and correction in CBCT
1. INTRODUCTION X-ray scatter in cone-beam CT (CBCT) is known to reduce image quality by introducing image artifacts, reducing constrast, and limiting CT number accuracy.1,2 The extent of the effect of x-ray scatter on CBCT image quality is determined by the shape and magnitude of the x-ray scatter distribution in the projection images. A method to allay the effects of scatter is imperative in enabling the applicability of CBCT to solve a wider domain of clinical problems. A method for correcting scatter using Monte Carlo (MC) simulations was previously outlined by Jarry et al.3 and involves estimating the scatter in each projection using a MC simulation consisting of a MC phantom of the object being imaged and MC model of the imaging geometry. The MC phantom can either have the density and material properties derived from a prior CT scan of the patient aligned to CBCT reconstruction being corrected (a scenario quite possible in image guided radiation therapy) or from the uncorrected CBCT reconstruction. The scatter estimations are then subtracted from the original projections to form a set of scatter corrected projections used to reconstruct a scatter free estimate of the CBCT reconstruction. This method as outlined in Jarry et al.3 required a significant amount of computational time to estimate the underlying scatter distribution, making its use in most clinical situations infeasible. In our previous work,4 the spectrum of the scatter distribution was investigated for a cylinder and two anthropomorphic phantoms. The results showed the spatial and angular frequencies of the scatter distribution vary depending on the imaging parameters (e.g., air gap, compensator), but in general are contained in the lower end of the frequency domain [spatial frequencies
F. Verhaegen Department of Radiation Oncology (MAASTRO), GROW—School for Oncology and Developmental Biology, Maastricht University Medical Center, Maastricht 6201 BN, The Netherthlands and Medical Physics Unit, Department of Oncology, McGill University, Montreal, Quebec H3G 1A4, Canada
D. A. Jaffray Radiation Medicine Program, Princess Margaret Cancer Centre, Toronto, Ontario M5G 2M9, Canada; Ontario Cancer Institute, Princess Margaret Cancer Centre, Toronto, Ontario M5G 2M9, Canada; and Department of Radiation Oncology, University of Toronto, Toronto, Ontario M5G 2M9, Canada
(Received 29 May 2014; revised 22 August 2014; accepted for publication 29 September 2014; published 16 December 2014) Purpose: X-ray scatter is a significant impediment to image quality improvements in cone-beam CT (CBCT). The authors present and demonstrate a novel scatter correction algorithm using a scatter estimation method that simultaneously combines multiple Monte Carlo (MC) CBCT simulations through the use of a concurrently evaluated fitting function, referred to as concurrent MC fitting (CMCF). Methods: The CMCF method uses concurrently run MC CBCT scatter projection simulations that are a subset of the projection angles used in the projection set, P, to be corrected. The scattered photons reaching the detector in each MC simulation are simultaneously aggregated by an algorithm which computes the scatter detector response, SMC. SMC is fit to a function, SF , and if the fit of SF is within a specified goodness of fit (GOF), the simulations are terminated. The fit, SF , is then used to interpolate the scatter distribution over all pixel locations for every projection angle in the set P. The CMCF algorithm was tested using a frequency limited sum of sines and cosines as the fitting function on both simulated and measured data. The simulated data consisted of an anthropomorphic head and a pelvis phantom created from CT data, simulated with and without the use of a compensator. The measured data were a pelvis scan of a phantom and patient taken on an Elekta Synergy platform. The simulated data were used to evaluate various GOF metrics as well as determine a suitable fitness value. The simulated data were also used to quantitatively evaluate the image quality improvements provided by the CMCF method. A qualitative analysis was performed on the measured data by comparing the CMCF scatter corrected reconstruction to the original uncorrected and corrected by a constant scatter correction reconstruction, as well as a reconstruction created using a set of projections taken with a small cone angle. Results: Pearson’s correlation, r, proved to be a suitable GOF metric with strong correlation with the actual error of the scatter fit, SF . Fitting the scatter distribution to a limited sum of sine and cosine functions using a low-pass filtered fast Fourier transform provided a computationally efficient and accurate fit. The CMCF algorithm reduces the number of photon histories required by over four orders of magnitude. The simulated experiments showed that using a compensator reduced the computational time by a factor between 1.5 and 1.75. The scatter estimates for the simulated and measured data were computed between 35–93 s and 114–122 s, respectively, using 16 Intel Xeon cores (3.0 GHz). The CMCF scatter correction improved the contrast-to-noise ratio by 10%–50% and reduced the reconstruction error to under 3% for the simulated phantoms. Conclusions: The novel CMCF algorithm significantly reduces the computation time required to estimate the scatter distribution by reducing the statistical noise in the MC scatter estimate and limiting the number of projection angles that must be simulated. Using the scatter estimate provided by the CMCF algorithm to correct both simulated and real projection data showed improved reconstruction image quality. C 2015 American Association of Physicists in Medicine. [http://dx.doi.org/10.1118/1.4903260] Key words: x-ray scatter, cone-beam CT, Monte Carlo, image quality
54
Med. Phys. 42 (1), January 2015
0094-2405/2015/42(1)/54/15/$30.00
© 2015 Am. Assoc. Phys. Med.
54
55
Bootsma, Verhaegen, and Jaffray: Efficient scatter distribution estimation and correction in CBCT
1. INTRODUCTION X-ray scatter in cone-beam CT (CBCT) is known to reduce image quality by introducing image artifacts, reducing constrast, and limiting CT number accuracy.1,2 The extent of the effect of x-ray scatter on CBCT image quality is determined by the shape and magnitude of the x-ray scatter distribution in the projection images. A method to allay the effects of scatter is imperative in enabling the applicability of CBCT to solve a wider domain of clinical problems. A method for correcting scatter using Monte Carlo (MC) simulations was previously outlined by Jarry et al.3 and involves estimating the scatter in each projection using a MC simulation consisting of a MC phantom of the object being imaged and MC model of the imaging geometry. The MC phantom can either have the density and material properties derived from a prior CT scan of the patient aligned to CBCT reconstruction being corrected (a scenario quite possible in image guided radiation therapy) or from the uncorrected CBCT reconstruction. The scatter estimations are then subtracted from the original projections to form a set of scatter corrected projections used to reconstruct a scatter free estimate of the CBCT reconstruction. This method as outlined in Jarry et al.3 required a significant amount of computational time to estimate the underlying scatter distribution, making its use in most clinical situations infeasible. In our previous work,4 the spectrum of the scatter distribution was investigated for a cylinder and two anthropomorphic phantoms. The results showed the spatial and angular frequencies of the scatter distribution vary depending on the imaging parameters (e.g., air gap, compensator), but in general are contained in the lower end of the frequency domain [spatial frequencies
