Yoshihiko Nagamine

GPU-based fast pencil beam algorithm for proton therapy

Performance of a treatment planning system is an essential factor in making sophisticated plans. The dose calculation is a major time-consuming process in planning operations. The standard algorithm for proton dose calculations is the pencil beam algorithm which produces relatively accurate results, but is time consuming. In order to shorten the computational time, we have developed a GPU (graphics processing unit)-based pencil beam algorithm. We have implemented this algorithm and calculated dose distributions in the case of a water phantom. The results were compared to those obtained by a traditional method with respect to the computational time and discrepancy between the two methods. The new algorithm shows 5-20 times faster performance using the NVIDIA GeForce GTX 480 card in comparison with the Intel Core-i7 920 processor. The maximum discrepancy of the dose distribution is within 0.2%. Our results show that GPUs are effective for proton dose calculations.

Biological effect of dose distortion by fiducial markers in spot-scanning proton therapy with a limited number of fields: A simulation study

PURPOSEnIn accurate proton spot-scanning therapy, continuous target tracking by fluoroscopic x ray during irradiation is beneficial not only for respiratory moving tumors of lung and liver but also for relatively stationary tumors of prostate. Implanted gold markers have been used with great effect for positioning the target volume by a fluoroscopy, especially for the cases of liver and prostate with the targets surrounded by water-equivalent tissues. However, recent studies have revealed that gold markers can cause a significant underdose in proton therapy. This paper focuses on prostate cancer and explores the possibility that multiple-field irradiation improves the underdose effect by markers on tumor-control probability (TCP).nnnMETHODSnA Monte Carlo simulation was performed to evaluate the dose distortion effect. A spherical gold marker was placed at several characteristic points in a water phantom. The markers were with two different diameters of 2 and 1.5 mm, both visible on fluoroscopy. Three beam arrangements of single-field uniform dose (SFUD) were examined: one lateral field, two opposite lateral fields, and three fields (two opposite lateral fields + anterior field). The relative biological effectiveness (RBE) was set to 1.1 and a dose of 74 Gy (RBE) was delivered to the target of a typical prostate size in 37 fractions. The ratios of TCP to that without the marker (TCP(r)) were compared with the parameters of the marker sizes, number of fields, and marker positions. To take into account the dependence of biological parameters in TCP model, α∕β values of 1.5, 3, and 10 Gy (RBE) were considered.nnnRESULTSnIt was found that the marker of 1.5 mm diameter does not affect the TCPs with all α∕β values when two or more fields are used. On the other hand, if the marker diameter is 2 mm, more than two irradiation fields are required to suppress the decrease in TCP from TCP(r) by less than 3%. This is especially true when multiple (two or three) markers are used for alignment of a patient.nnnCONCLUSIONSnIt is recommended that 1.5-mm markers be used to avoid the reduction of TCP as well as to spare the surrounding critical organs, as long as the markers are visible on x-ray fluoroscopy. When 2-mm markers are implanted, more than two fields should be used and the markers should not be placed close to the distal edge of any of the beams.

Evaluation of the influence of double and triple Gaussian proton kernel models on accuracy of dose calculations for spot scanning technique.

PURPOSEnThe main purpose in this study was to present the results of beam modeling and how the authors systematically investigated the influence of double and triple Gaussian proton kernel models on the accuracy of dose calculations for spot scanning technique.nnnMETHODSnThe accuracy of calculations was important for treatment planning software (TPS) because the energy, spot position, and absolute dose had to be determined by TPS for the spot scanning technique. The dose distribution was calculated by convolving in-air fluence with the dose kernel. The dose kernel was the in-water 3D dose distribution of an infinitesimal pencil beam and consisted of an integral depth dose (IDD) and a lateral distribution. Accurate modeling of the low-dose region was important for spot scanning technique because the dose distribution was formed by cumulating hundreds or thousands of delivered beams. The authors employed a double Gaussian function as the in-air fluence model of an individual beam. Double and triple Gaussian kernel models were also prepared for comparison. The parameters of the kernel lateral model were derived by fitting a simulated in-water lateral dose profile induced by an infinitesimal proton beam, whose emittance was zero, at various depths using Monte Carlo (MC) simulation. The fitted parameters were interpolated as a function of depth in water and stored as a separate look-up table. These stored parameters for each energy and depth in water were acquired from the look-up table when incorporating them into the TPS. The modeling process for the in-air fluence and IDD was based on the method proposed in the literature. These were derived using MC simulation and measured data. The authors compared the measured and calculated absolute doses at the center of the spread-out Bragg peak (SOBP) under various volumetric irradiation conditions to systematically investigate the influence of the two types of kernel models on the dose calculations.nnnRESULTSnThe authors investigated the difference between double and triple Gaussian kernel models. The authors found that the difference between the two studied kernel models appeared at mid-depths and the accuracy of predicting the double Gaussian model deteriorated at the low-dose bump that appeared at mid-depths. When the authors employed the double Gaussian kernel model, the accuracy of calculations for the absolute dose at the center of the SOBP varied with irradiation conditions and the maximum difference was 3.4%. In contrast, the results obtained from calculations with the triple Gaussian kernel model indicated good agreement with the measurements within ±1.1%, regardless of the irradiation conditions.nnnCONCLUSIONSnThe difference between the results obtained with the two types of studied kernel models was distinct in the high energy region. The accuracy of calculations with the double Gaussian kernel model varied with the field size and SOBP width because the accuracy of prediction with the double Gaussian model was insufficient at the low-dose bump. The evaluation was only qualitative under limited volumetric irradiation conditions. Further accumulation of measured data would be needed to quantitatively comprehend what influence the double and triple Gaussian kernel models had on the accuracy of dose calculations.

A patient‐specific aperture system with an energy absorber for spot scanning proton beams: Verification for clinical application

PURPOSEnIn the authors proton therapy system, the patient-specific aperture can be attached to the nozzle of spot scanning beams to shape an irradiation field and reduce lateral fall-off. The authors herein verified this system for clinical application.nnnMETHODSnThe authors prepared four types of patient-specific aperture systems equipped with an energy absorber to irradiate shallow regions less than 4 g/cm(2). The aperture was made of 3-cm-thick brass and the maximum water equivalent penetration to be used with this system was estimated to be 15 g/cm(2). The authors measured in-air lateral profiles at the isocenter plane and integral depth doses with the energy absorber. All input data were obtained by the Monte Carlo calculation, and its parameters were tuned to reproduce measurements. The fluence of single spots in water was modeled as a triple Gaussian function and the dose distribution was calculated using a fluence dose model. The authors compared in-air and in-water lateral profiles and depth doses between calculations and measurements for various apertures of square, half, and U-shaped fields. The absolute doses and dose distributions with the aperture were then validated by patient-specific quality assurance. Measured data were obtained by various chambers and a 2D ion chamber detector array.nnnRESULTSnThe patient-specific aperture reduced the penumbra from 30% to 70%, for example, from 34.0 to 23.6 mm and 18.8 to 5.6 mm. The calculated field width for square-shaped apertures agreed with measurements within 1 mm. Regarding patient-specific aperture plans, calculated and measured doses agreed within -0.06% ± 0.63% (mean ± SD) and 97.1% points passed the 2%-dose/2 mm-distance criteria of the γ-index on average.nnnCONCLUSIONSnThe patient-specific aperture system improved dose distributions, particularly in shallow-region plans.

TH-C-BRD-04: Beam Modeling and Validation with Triple and Double Gaussian Dose Kernel for Spot Scanning Proton Beams

PURPOSEnTo present the validity of our beam modeling with double and triple Gaussian dose kernels for spot scanning proton beams in Nagoya Proton Therapy Center. This study investigates the conformance between the measurements and calculation results in absolute dose with two types of beam kernel.nnnMETHODSnA dose kernel is one of the important input data required for the treatment planning software. The dose kernel is the 3D dose distribution of an infinitesimal pencil beam of protons in water and consists of integral depth doses and lateral distributions. We have adopted double and triple Gaussian model as lateral distribution in order to take account of the large angle scattering due to nuclear reaction by fitting simulated inwater lateral dose profile for needle proton beam at various depths. The fitted parameters were interpolated as a function of depth in water and were stored as a separate look-up table for the each beam energy. The process of beam modeling is based on the method of MDACC [X.R.Zhu 2013].nnnRESULTSnFrom the comparison results between the absolute doses calculated by double Gaussian model and those measured at the center of SOBP, the difference is increased up to 3.5% in the high-energy region because the large angle scattering due to nuclear reaction is not sufficiently considered at intermediate depths in the double Gaussian model. In case of employing triple Gaussian dose kernels, the measured absolute dose at the center of SOBP agrees with calculation within ±1% regardless of the SOBP width and maximum range.nnnCONCLUSIONnWe have demonstrated the beam modeling results of dose distribution employing double and triple Gaussian dose kernel. Treatment planning system with the triple Gaussian dose kernel has been successfully verified and applied to the patient treatment with a spot scanning technique in Nagoya Proton Therapy Center.

SU-E-CAMPUS-J-03: Commissioning of the On-Board Cone-Beam CT System Equipped On the Rotating Gantry of a Proton Therapy System

PURPOSEnProton therapy requires highly-precise image guidance in patient setup to ensure accurate dose delivery. Cone-beam CT (CBCT) is expected to play an important role to reduce uncertainties in patient setup. Hokkaido University has developed a new proton therapy system dedicated to spot-scanning under a collaborative work with Hitachi Ltd. In our system, an orthogonal X-ray imaging system is mounted on a full-rotating gantry. On-board CBCT imaging is therefore available. We have conducted commissioning of the CBCT system for clinical use in proton therapy.nnnMETHODSnThe orthogonal X-ray imaging system, which consists of two sets of X-ray tubes and flat panel detectors (FPDs), are equipped on the rotating gantry. The FPDs are mounted on the proton beam nozzle and can be retracted when not in use. The distance between the X-ray source and the FPD is about 2.1 m. The maximum rotation speed of the gantry is 1 rpm, so CBCT images can be acquired in approximately 1 minute. The maximum reconstruction volume is nearly 40 cm in diameter and 20 cm in axial length. For commissioning of the CBCT system, mechanical accuracy of the rotating gantry first was evaluated. Imaging performance was examined via quantitative evaluation of image quality.nnnRESULTSnThrough the mechanical test, the isocentricity of the gantry was confirmed to be less than 1 mm. Moreover, it was improved to 0.5 mm with an appropriate correction. The accurate rotation of the gantry contributes to the CBCT image quality. In the image quality test, objects with 7 line-pairs per cm, which corresponds to a line spacing of 0.071 cm, could be discerned. Spatial linearity and uniformity were also sufficient.nnnCONCLUSIONnClinical commissioning of the on-board CBCT system for proton therapy was conducted, and CBCT images with sufficient quality were successfully obtained. This research was supported by the Cabinet Office, Government of Japan and the Japan Society for the Promotion of Science (JSPS) through the Funding Program for World-Leading Innovative R&D on Science and Technology (FIRST Program), initiated by the Council for Science and Technology Policy (CSTP).

MO‐F‐213AB‐04: Biological Effect of Dose Shadowing by Fiducial Markers in Spot Scanning Proton Therapy with a Limited Number of Fields

PURPOSEnIn spot scanning proton therapy, accurate patient positioning before and during treatment is essential. A small gold ball marker is suitable as a fiducial for prostate treatment. However, it has been pointed out that the marker causes dose shadowing because the protons are scattered with their energy quickly diminished. In this research we explore the possibility that the biological effect of dose shadowing can be mitigated with a limited number of fields.nnnMETHODSnThe proton dose distribution in prostate was simulated using Geant4. The simulations include the Hokkaido University spot scanning nozzle and a water phantom positioned isocentrically. The PTV was delineated at the center of the phantom and a gold ball of 2 mm in diameter was placed at the middle of the PTV. The plan was created by single-field optimization and each of the following beam arrangements was investigated; (1) single lateral field (2) two lateral fields (3) two lateral + one anterior fields (4) four-field box. The dose prescription was D95 = 74 GyE (37 fr). The minimum dose and tumor control probability (TCP) were compared for the four beam arrangements.nnnRESULTSnFor (1)-(4), the minimum dose values were 55%, 77%, 78%, and 84% of the prescribed dose, respectively. The reduction of the TCP values from those in the absence of the gold marker were 50%, 2%, 1.1%, and 0.7%, using the TCP model by Wang et al. (Int.J.Radiat.Oncol.Biol.Phys. 55, 2003) and 2%, 0.7%, 0.5%, and 0.4%, using the biological parameters in Levegrün et al. (Int.J.RadiatOncol.Biol.Phys. 51, 2001), respectively.nnnCONCLUSIONSnAlthough dose shadowing by the gold marker is locally non-negligible, the size of the affected domain is tiny. It was found that with a minimum number of fields, the TCP nearly recovers to the value without the gold marker.

Efficient registration method of medical images using GPU

Registration of medical images is an important task; however, automatic image-based registration is computationally expensive. Given this task, the authors propose an efficient rigid registration method, which is based on mutual information and uses a graphics processing unit (GPU). Mutual-information-based registration methods require joint-histogram computation. Although a GPU can provide high performance computing, a joint histogram has a large number of bins, and the computation of such a histogram is not suitable for a GPU (whose shared memory is limited). Taking advantage of the fact that one image (the reference image) is not transformed during the registration process, the proposed method computes a joint histogram by computing multiple onedimensional histograms and combining them. The method can therefore be efficiently implemented on a GPU even with limited shared memory. Experimental results for 256 × 256 × 256 image registration show that the method is about 140 times faster than a standard implementation on a CPU and 2.6 times faster than previous methods using GPUs.