Dose Calculation Algorithms for External Radiation Therapy: An Overview for Practitioners

A deep learning based dynamic arc radiotherapy photon dose engine trained on Monte Carlo dose distributions

Background and purpose

Despite hardware acceleration, state-of-the-art Monte Carlo (MC) dose engines require considerable computation time to reduce stochastic noise. We developed a deep learning (DL) based dose engine reaching high accuracy at strongly reduced computation times.

Materials and methods

Radiotherapy treatment plans and computed tomography scans were collected for 350 treatments in a variety of tumor sites. Dose distributions were computed using a MC dose engine for ∼ 30,000 separate segments at 6 MV and 10 MV beam energies, both flattened and flattening filter free. For dynamic arcs these explicitly incorporated the leaf, jaw and gantry motions during dose delivery. A neural network was developed, combining two-dimensional convolution and recurrence using 64 hidden channels. Parameters were trained to minimize the mean squared log error loss between the MC computed dose and the model output. Full dose distributions were reconstructed for 100 additional treatment plans. Gamma analyses were performed to assess accuracy.

Results

DL dose evaluation was on average 82 times faster than MC computation at a 1 % accuracy setting. In voxels receiving at least 10 % of the maximum dose the overall global gamma pass rate using a 2 % and 2 mm criterion was 99.6 %, while mean local gamma values were accurate within 2 %. In the high dose region over 50 % of maximum the mean local gamma approached a 1 % accuracy.

Conclusions

A DL based dose engine was implemented, able to accurately reproduce MC computed dynamic arc radiotherapy dose distributions at high speed.

Evaluation of monte carlo to support commissioning of the treatment planning system of new pencil beam scanning proton therapy facilities

Objective. To demonstrate the potential of Monte Carlo (MC) to support the resource-intensive measurements that comprise the commissioning of the treatment planning system (TPS) of new proton therapy facilities.Approach. Beam models of a pencil beam scanning system (Varian ProBeam) were developed in GATE (v8.2), Eclipse proton convolution superposition algorithm (v16.1, Varian Medical Systems) and RayStation MC (v12.0.100.0, RaySearch Laboratories), using the beam commissioning data. All models were first benchmarked against the same commissioning data and validated on seven spread-out Bragg peak (SOBP) plans. Then, we explored the use of MC to optimise dose calculation parameters, fully understand the performance and limitations of TPS in homogeneous fields and support the development of patient-specific quality assurance (PSQA) processes. We compared the dose calculations of the TPSs against measurements (DDTPSvs.Meas.) or GATE (DDTPSvs.GATE) for an extensive set of plans of varying complexity. This included homogeneous plans with varying field-size, range, width, and range-shifters (RSs) (n= 46) and PSQA plans for different anatomical sites (n= 11).Main results. The three beam models showed good agreement against the commissioning data, and dose differences of 3.5% and 5% were found for SOBP plans without and with RSs, respectively. DDTPSvs.Meas.and DDTPSvs.GATEwere correlated in most scenarios. In homogeneous fields the Pearson's correlation coefficient was 0.92 and 0.68 for Eclipse and RayStation, respectively. The standard deviation of the differences between GATE and measurements (±0.5% for homogeneous and ±0.8% for PSQA plans) was applied as tolerance when comparing TPSs with GATE. 72% and 60% of the plans were within the GATE predicted dose difference for both TPSs, for homogeneous and PSQA cases, respectively.Significance. Developing and validating a MC beam model early on into the commissioning of new proton therapy facilities can support the validation of the TPS and facilitate comprehensive investigation of its capabilities and limitations.

Carbon ion radiography with a composite ionization chamber detector

Range uncertainty in carbon ion therapy can diminish treatment efficacy because it may cause deviation from the planned dose distribution. The precise and accurate determination of relative stopping power (RSP) maps of carbon ions in the patient is a direct solution to this problem. To obtain RSP maps in patients undergoing carbon ion radiography, our team developed a preliminary prototype of a composite ionization chamber detection (CICD) system. The CICD prototype employs synchronously gated integral electronics with the ability to measure the depth-to-dose curve and the beam profile simultaneously. Carbon ion radiography experiments were performed on hemispherical, sloped, and stepped phantoms using the Heavy Ion Medical Machine (HIMM) beam. The beam energy was 190.19 MeV/μ and the beam spot full width at half maximum (FWHM) was 7.42 mm. The radiographic image of the sloped phantom, the thickness prediction accuracy of each pixel (2 mm) is 88.25%, its absolute mean error (AME) is 1.07 mm, and the maximum absolute deviation (MAD) is 2.64 mm. The prediction accuracy of the CICD prototype is mainly affected by electronic noise, with a noise-to-signal ratio (NSR) of about 14.36 dB. Carbon ion radiography simulations were performed in this study using Geant4 software to eliminate the effect of the electronic noise. The thickness prediction accuracy is 98.54%, 98.62%, and 99.07% per pixel for hemispherical, sloped and stepped phantoms, respectively, with AME of 0.09 mm, 0.27 mm, and 0.48 mm. Carbon ion radiography utilizing the CICD prototype scheme has the ability to refine the accuracy and resolution of radiographic images, consequently establishing a scientific foundation for diminishing the effects of range uncertainty and fully exploiting the advantages of precision particle therapy.

A systematic approach for calibrating a Monte Carlo code to a treatment planning system for obtaining dose, LET, variable proton RBE and out-of-field dose

Objective. While integration of variable relative biological effectiveness (RBE) has not reached full clinical implementation, the importance of having the ability to recalculate proton treatment plans in a flexible, dedicated Monte Carlo (MC) code cannot be understated . Here we provide a step-wise method for calibrating dose from a MC code to a treatment planning system (TPS), to obtain required parameters for calculating linear energy transfer (LET), variable RBE and in general enabling clinical realistic research studies beyond the capabilities of a TPS.Approach. Initially, Pristine Bragg peaks (PBP) were calculated in both the Eclipse TPS and the FLUKA MC code. A rearranged Bortfeld energy-range relation was applied to the initial energy of the beam to fine-tune the range of the MC code at 80% dose level distal to the PBP. The energy spread was adapted by dividing the TPS range by the MC range for dose level 80%-20% distal to the PBP. Density and relative proton stopping power were adjusted by comparing the TPS and MC for different Hounsfield units. To find the relationship of dose per primary particle from the MC to dose per monitor unit in the TPS, integration was applied to the area of the Bragg curve. The calibration was validated for spread-out Bragg peaks (SOBP) in water and patient treatment plans. Following the validation, variable RBE were calculated using established models.Main results.The PBPs ranges were within ±0.3mm threshold, and a maximum of 5.5% difference for the SOBPs was observed. The patient validation showed excellent dose agreement between the TPS and MC, with the greatest differences for the lung tumor patient.Significance. Aprocedure for calibrating a MC code to a TPS was developed and validated. The procedure enables MC-based calculation of dose, LET, variable RBE, advanced (secondary) particle tracking and more from treatment plans.

Effects of tumour heterogeneous properties on modelling the transport of radiative particles

Dose calculation plays a critical role in radiotherapy (RT) treatment planning, and there is a growing need to develop accurate dose deposition models that incorporate heterogeneous tumour properties. Deterministic models have demonstrated their capability in this regard, making them the focus of recent treatment planning studies as they serve as a basis for simplified models in RT treatment planning. In this study, we present a simplified deterministic model for photon transport based on the Boltzmann transport equation (BTE) as a proof-of-concept to illustrate the impact of heterogeneous tumour properties on RT treatment planning. We employ the finite element method (FEM) to simulate the photon flux and dose deposition in real cases of diffuse intrinsic pontine glioma (DIPG) and neuroblastoma (NB) tumours. Importantly, in light of the availability of pipelines capable of extracting tumour properties from magnetic resonance imaging (MRI) data, we highlight the significance of such data. Specifically, we utilise cellularity data extracted from DIPG and NB MRI images to demonstrate the importance of heterogeneity in dose calculation. Our model simplifies the process of simulating a RT treatment system and can serve as a useful starting point for further research. To simulate a full RT treatment system, one would need a comprehensive model that couples the transport of electrons and photons.

Fast dose calculation in x-ray guided interventions by using deep learning

Objective.Patient dose estimation in x-ray-guided interventions is essential to prevent radiation-induced biological side effects. Current dose monitoring systems estimate the skin dose based in dose metrics such as the reference air kerma. However, these approximations do not take into account the exact patient morphology and organs composition. Furthermore, accurate organ dose estimation has not been proposed for these procedures. Monte Carlo simulation can accurately estimate the dose by recreating the irradiation process generated during the x-ray imaging, but at a high computation time, limiting an intra-operative application. This work presents a fast deep convolutional neural network trained with MC simulations for patient dose estimation during x-ray-guided interventions.Approach.We introduced a modified 3D U-Net that utilizes a patient's CT scan and the numerical values of imaging settings as input to produce a Monte Carlo dose map. To create a dataset of dose maps, we simulated the x-ray irradiation process for the abdominal region using a publicly available dataset of 82 patient CT scans. The simulation involved varying the angulation, position, and tube voltage of the x-ray source for each scan. We additionally conducted a clinical study during endovascular abdominal aortic repairs to validate the reliability of our Monte Carlo simulation dose maps. Dose measurements were taken at four specific anatomical points on the skin and compared to the corresponding simulated doses. The proposed network was trained using a 4-fold cross-validation approach with 65 patients, and evaluating the performance on the remaining 17 patients during testing.Main results.The clinical validation demonstrated a average error within the anatomical points of 5.1%. The network yielded test errors of 11.5 ± 4.6% and 6.2 ± 1.5% for peak and average skin doses, respectively. Furthermore, the mean errors for the abdominal region and pancreas doses were 5.0 ± 1.4% and 13.1 ± 2.7%, respectively.Significance.Our network can accurately predict a personalized 3D dose map considering the current imaging settings. A short computation time was achieved, making our approach a potential solution for dose monitoring and reporting commercial systems.

Analysis of Performance and Failure Modes of the IROC Proton Liver Phantom

Purpose

To analyze trends in institutional performance and failure modes for the Imaging and Radiation Oncology Core's (IROC's) proton liver phantom.

Materials and Methods

Results of 66 phantom irradiations from 28 institutions between 2015 and 2020 were retrospectively analyzed. Univariate analysis and random forest models were used to associate irradiation conditions with phantom results. Phantom results included pass/fail classification, average thermoluminescent dosimeter (TLD) ratio of both targets, and percentage of pixels passing gamma of both targets. The following categories were evaluated in terms of how they predicted these outcomes: irradiation year, treatment planning system (TPS), TPS algorithm, treatment machine, number of irradiations, treatment technique, motion management technique, number of isocenters, and superior-inferior extent (in cm) of the 90% TPS isodose line for primary target 1 (PTV1) and primary target 2 (PTV2). In addition, failures were categorized by failure mode.

Results

Average pass rate was approximately 52% and average TLD ratio for both targets had slightly improved. As the treatment field increased to cover the target, the pass rate statistically significantly fell. Lower pass rates were observed for Mevion machines, scattered irradiation techniques, and gating and internal target volume (ITV) motion management techniques. Overall, the accuracy of the random forest modeling of the phantom results was approximately 73% ± 14%. The most important predictor was the superior-inferior extent for both targets and irradiation year. Three failure modes dominated the failures of the phantom: (1) systematic underdosing, (2) poor localization in the superior-inferior direction, and (3) range error. Only 44% of failures have similar failure modes between the 2 targets.

Conclusion

Improvement of the proton liver phantom has been observed; however, the pass rate remains the lowest among all IROC phantoms. Through various analysis techniques, range uncertainty, motion management, and underdosing are the main culprits of failures of the proton liver phantom. Clinically, careful consideration of the influences of liver proton therapy is needed to improve phantom performance and patient outcome.

A deep-learning-based dose verification tool utilizing fluence maps for a cobalt-60 compensator-based intensity-modulated radiation therapy system

Background and purpose

A novel cobalt-60 compensator-based intensity-modulated radiation therapy (IMRT) system was developed for a resource-limited environment but lacked an efficient dose verification algorithm. The aim of this study was to develop a deep-learning-based dose verification algorithm for accurate and rapid dose predictions.

Materials and methods

A deep-learning network was employed to predict the doses from static fields related to beam commissioning. Inputs were a cube-shaped phantom, a beam binary mask, and an intersecting volume of the phantom and beam binary mask, while output was a 3-dimensional (3D) dose. The same network was extended to predict patient-specific doses for head and neck cancers using two different approaches. A field-based method predicted doses for each field and combined all calculated doses into a plan, while the plan-based method combined all nine fluences into a plan to predict doses. Inputs included patient computed tomography (CT) scans, binary beam masks, and fluence maps truncated to the patient's CT in 3D.

Results

For static fields, predictions agreed well with ground truths with average deviations of less than 0.5% for percent depth doses and profiles. Even though the field-based method showed excellent prediction performance for each field, the plan-based method showed better agreement between clinical and predicted dose distributions. The distributed dose deviations for all planned target volumes and organs at risk were within 1.3 Gy. The calculation speed for each case was within two seconds.

Conclusions

A deep-learning-based dose verification tool can accurately and rapidly predict doses for a novel cobalt-60 compensator-based IMRT system.

Feasibility of a multigroup Boltzmann-Fokker-Planck solution for electron beam dose calculations

Legacy nuclear-reactor Boltzmann solvers start clinical deployment as an alternative to Monte Carlo (MC) codes and Fermi-Eyges semiemprical models in radiation oncology treatment planning. Today's certified clinical solvers are limited to photon beams. In this paper, ELECTR, a state-of-the-art multigroup electron cross sections generation module in NJOY is presented and validated against Lockwood's calorimetric measurements, EGS-nrc and GEANT-4 for 1-20 MeV unidirectional electron beams. The nuclear-reactor DRAGON-5 solver is upgraded to access the library and solve the Boltzmann-Fokker-Planck (BFP) equation. A variety of heterogeneous radiotherapy and radiosurgery phantom configurations were used for validation purpose. Case studies include a thorax benchmark, that of a typical breast Intra-Operative Radiotherapy and a high-heterogeneity patient-like benchmark. For all beams, [Formula: see text] of the water voxels satisfied the American Association of Physicists in Medicine accuracy criterion for a BFP-MC dose error below [Formula: see text]. At least, [Formula: see text] of adipose, muscle, bone, lung, tumor and breast voxels satisfied the [Formula: see text] criterion. The average BFP-MC relative error was about [Formula: see text] for all voxels, beams and materials combined. By irradiating homogeneous slabs from [Formula: see text] (hydrogen) to [Formula: see text] (einsteinium), we reported performance and defects of the CEPXS mode [US. Sandia National Lab., SAND-89-1685] in ELECTR for the entire periodic table. For all Lockwood's benchmarks, NJOY-DRAGON dose predictions are within the experimental data precision for [Formula: see text] of voxels.

Exposure of the heart in lung cancer radiation therapy: A systematic review of heart doses published during 2013 to 2020

BACKGROUND AND PURPOSE

Lung cancer radiotherapy increases the risk of cardiotoxicity and heart radiation dose is an independent predictor of poor survival. This study describes heart doses and strategies aiming to reduce exposure.

MATERIALS AND METHODS

A systematic review of lung cancer dosimetry studies reporting heart doses published 2013-2020 was undertaken. Doses were compared according to laterality, region irradiated, treatment modality (stereotactic ablative body radiotherapy (SABR) and non-SABR), planning technique, and respiratory motion management.

RESULTS

For 392 non-SABR regimens in 105 studies, the average MHD was 10.3 Gy (0.0-48.4) and was not significantly different between left and right-sided tumours. It was similar between IMRT and 3DCRT (10.9 Gy versus 10.6 Gy) and lower with particle beam therapy (proton 7.0 Gy; carbon-ion 1.9 Gy). Active respiratory motion management reduced exposure (7.4 Gy versus 9.3 Gy). For 168 SABR regimens in 35 studies, MHD was 4.0 Gy (0.0-32.4). Exposure was higher in central and lower lobe lesions (6.3 and 5.8 Gy respectively). MHD was lowest for carbon ions (0.5 Gy) compared to other techniques. Active respiratory motion management reduced exposure (2.4 Gy versus 5.0 Gy). Delineation guidelines and Dose Volume Constraints for the heart varied substantially.

CONCLUSIONS

There is scope to reduce heart radiation dose in lung cancer radiotherapy. Consensus on planning objectives, contouring and DVCs for the heart may lead to reduced heart doses in the future. For IMRT, more stringent optimisation objectives may reduce heart dose. Active respiratory motion management or particle therapy may be considered in situations where cardiac dose is high.