Monte Carlo Dose Calculation Using MRI Based Synthetic CT Generated by Fully Convolutional Neural Network for Gamma Knife Radiosurgery

The aim of this work is to study the dosimetric effect from generated synthetic computed tomography (sCT) from magnetic resonance (MR) images using a deep learning algorithm for Gamma Knife (GK) stereotactic radiosurgery (SRS). The Monte Carlo (MC) method is used for dose calculations. Thirty patients were retrospectively selected with our institution IRB's approval. All patients were treated with GK SRS based on T1-weighted MR images and also underwent conventional external beam treatment with a CT scan. Image datasets were preprocessed with registration and were normalized to obtain similar intensity for the pairs of MR and CT images. A deep convolutional neural network arranged in an encoder-decoder fashion was used to learn the direct mapping from MR to the corresponding CT. A number of metrics including the voxel-wise mean error (ME) and mean absolute error (MAE) were used for evaluating the difference between generated sCT and the true CT. To study the dosimetric accuracy, MC simulations were performed based on the true CT and sCT using the same treatment parameters. The method produced an MAE of 86.6 ± 34.1 Hundsfield units (HU) and a mean squared error (MSE) of 160.9 ± 32.8. The mean Dice similarity coefficient was 0.82 ± 0.05 for HU > 200. The difference for dose-volume parameter D95 between the ground true dose and the dose calculated with sCT was 1.1% if a synthetic CT-to-density table was used, and 4.9% compared with the calculations based on the water-brain phantom.

A Monte Carlo model and its commissioning for the Leksell Gamma Knife Perfexion radiosurgery system

PURPOSE

To develop and commission a Monte Carlo (MC) simulation model for the Leksell Gamma Knife (LGK) Perfexion (PFX) radiosurgery system.

METHOD

We previously established a source model for MC simulations of the LGK PFX for the purpose of the treatment planning system (TPS) dose verification and plan evaluation. To make practical and effective use of the model in clinic, several issues need to be addressed. First, thorough commissioning procedures are needed to ensure the validity of the model parameters, such as the source-to-focus (STF) distance, the source solid angle. Second, an efficient source particle sampling method is required to facilitate dose calculations for multitarget and multishot configurations in patient treatment plans. Third, inseparably, it is interesting to know the dose difference between the two GK TPS algorithms (TMR and convolution) and the MC method in extreme heterogeneous cases resulting from the inhomogeneous effect. We report our recent development in addressing these issues. Phantoms with the frame fiducials were manually created in the format of DICOM CT image to eliminate the uncertainties associated with scanner artifacts and image registration. The created homogeneous phantom was used to calibrate the model parameters to match the output factors with the manufacturer provided data, and the heterogeneous phantom with multilayer materials was used to study the inhomogeneous effect.

RESULTS

The agreement between the MC calculation and TPS was very good for the homogeneous spherical phantom. The difference of the full width at half maximum (FWHM) of the profiles was less than 1 mm except for the profile for 16 mm collimator along z-axis (less than 2 mm). For the extreme heterogeneous test case, it was shown that the TMR algorithm can overestimate the target dose by up to 22% using the measure of dose volume parameter D95. The agreement between the MC method and the TPS convolution method was better (within 3.6%) for the target near the center of phantom, however, discrepancy (up to 10.7%) existed for the target close to the skull. The difference between the two TPS dose algorithms was about 11%.

CONCLUSIONS

Considerable dose difference may result from the effect of heterogeneity, such as in the regions of the air cavities and bones. As the MC method has been extensively used in conventional external beams, it is worthwhile for further investigation in applying the MC method to accurate dose planning in the new GK PFX radiosurgery platform.

On the development of a comprehensive MC simulation model for the Gamma Knife Perfexion radiosurgery unit

This work presents a comprehensive Monte Carlo (MC) simulation model for the Gamma Knife Perfexion (PFX) radiosurgery unit. Model-based dosimetry calculations were benchmarked in terms of relative dose profiles (RDPs) and output factors (OFs), against corresponding EBT2 measurements. To reduce the rather prolonged computational time associated with the comprehensive PFX model MC simulations, two approximations were explored and evaluated on the grounds of dosimetric accuracy. The first consists in directional biasing of the (60)Co photon emission while the second refers to the implementation of simplified source geometric models. The effect of the dose scoring volume dimensions in OF calculations accuracy was also explored. RDP calculations for the comprehensive PFX model were found to be in agreement with corresponding EBT2 measurements. Output factors of 0.819  ±  0.004 and 0.8941  ±  0.0013 were calculated for the 4 mm and 8 mm collimator, respectively, which agree, within uncertainties, with corresponding EBT2 measurements and published experimental data. Volume averaging was found to affect OF results by more than 0.3% for scoring volume radii greater than 0.5 mm and 1.4 mm for the 4 mm and 8 mm collimators, respectively. Directional biasing of photon emission resulted in a time efficiency gain factor of up to 210 with respect to the isotropic photon emission. Although no considerable effect on relative dose profiles was detected, directional biasing led to OF overestimations which were more pronounced for the 4 mm collimator and increased with decreasing emission cone half-angle, reaching up to 6% for a 5° angle. Implementation of simplified source models revealed that omitting the sources' stainless steel capsule significantly affects both OF results and relative dose profiles, while the aluminum-based bushing did not exhibit considerable dosimetric effect. In conclusion, the results of this work suggest that any PFX simulation model should be benchmarked in terms of both RDP and OF results.

Patient-specific independent 3D GammaPlan quality assurance for Gamma Knife Perfexion radiosurgery

One of the most important aspects of quality assurance (QA) in radiation therapy is redundancy of patient treatment dose calculation. This work is focused on the patient-specific time and 3D dose treatment plan verification for stereotactic radiosurgery using Leksell Gamma Knife Perfexion (LGK PFX). The virtual model of LGK PFX was developed in MATLAB, based on the physical dimensions provided by the manufacturer. The ring-specific linear attenuation coefficients (LAC) and output factors (OFs) reported by the manufacturer were replaced by the measurement-based collimator size-specific OFs and a single LAC = 0.0065 mm-1. Calculation depths for each LGK PFX shot were obtained by ray-tracing technique, and the dose calculation formalism was similar to the one used by GammaPlan treatment planning software versions 8 and 9. The architecture of the QA process was based on the in-house online database search of the LGK PFX database search for plan-specific information. A series of QA phantom plans was examined to verify geometric and dosimetric accuracy of the software. The accuracy of the QA process was further evaluated through evaluation of a series of patient plans. The shot time/focus point dose verification for each shot took less than 1 sec/shot with full 3D isodose verification taking about 30 sec/shot on a desktop PC. GammaPlan database access time took less than 0.05 sec. The geometric accuracy (location of the point of maximum dose) of the phantom and patient plan was dependent on the resolution of the original dose matrix and was of the order of 1 dose element. Dosimetric accuracy of the independently calculated phantom and patient point (focus) doses was within 3.5% from the GammaPlan, with the mean = 2.3% and SD= 1.1%. The process for independent pretreatment patient-specific Gamma Knife Perfexion time and dose verification was created and validated.