Millisecond speed deep learning based proton dose calculation with Monte Carlo accuracy

Pastor-Serrano, Oscar; Perkó, Zoltán

doi:10.1088/1361-6560/ac692e

Cited by 27 publications

(27 citation statements)

References 70 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It was recently shown that the transformer architecture is very suitable for dose deposition predictions in the case of the proton beam therapy 17,19 . Transformer models rely on the prediction of sequences to sequences utilizing the so‐called attention mechanism 18 .…”

Section: Methodsmentioning

confidence: 99%

“…17,19 Transformer models rely on the prediction of sequences to sequences utilizing the so-called attention mechanism. 18 In the case of the model of interest, named DoTA, 17,19 the source sequence comprises density matrix tokens, which were obtained from encoding the phantom tissue density matrix slice by slice using a convolutional encoder. The target sequence comprises energy deposition tokens which are decoded via a convolutional decoder into energy deposition slices.…”

Section: Transformer-based Modelmentioning

confidence: 99%

“…It was recently shown that the transformer architecture is very suitable for dose deposition predictions in the case of the proton beam therapy. 17,19 Transformer models rely on the prediction of sequences to sequences utilizing the so-called attention mechanism. 18 In the case of the model of interest, named DoTA, 17,19 the source sequence comprises density matrix tokens, which were obtained from encoding the phantom tissue density matrix slice by slice using a convolutional encoder.…”

Section: Transformer-based Modelmentioning

confidence: 99%

“…15 The two 3D U-Net models differ in the way they are trained. The third model comprises a transformer architecture recently proposed for dose deposition predictions in the proton therapy 19 based on the self-attention mechanism. 18…”

Section: Machine Learning Modelsmentioning

confidence: 99%

“…This approach may be more flexible than a volumetric prediction as stacking of different tissue slices is a very general and potentially easier transferable approach to novel geometries than predicting the entire 3D phantom geometry at once. 19 This presents, however, a trade-off between the complexity of the model, whose increase usually requires more training data, and the ease with which a volume is being rendered. In the case of the 3D U-Net model, the structure is relatively simple, as it contains a series of downsampling convolutions followed by a number of upsampling convolutions, which are connected at the same level with skip connections.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Small beams, fast predictions: a comparison of machine learning dose prediction models for proton minibeam therapy

et al. 2022

View full text Add to dashboard Cite

Background Dose calculations for novel radiotherapy cancer treatments such as proton minibeam radiation therapy is often done using full Monte Carlo (MC) simulations. As MC simulations can be very time consuming for this kind of application, deep learning models have been considered to accelerate dose estimation in cancer patients. Purpose This work systematically evaluates the dose prediction accuracy, speed and generalization performance of three selected state‐of‐the‐art deep learning models for dose prediction applied to the proton minibeam therapy. The strengths and weaknesses of those models are thoroughly investigated, helping other researchers to decide on a viable algorithm for their own application. Methods The following recently published models are compared: first, a 3D U‐Net model trained as a regression network, second, a 3D U‐Net trained as a generator of a generative adversarial network (GAN) and third, a dose transformer model which interprets the dose prediction as a sequence translation task. These models are trained to emulate the result of MC simulations. The dose depositions of a proton minibeam with a diameter of 800μm and an energy of 20–100 MeV inside a simple head phantom calculated by full Geant4 MC simulations are used as a case study for this comparison. The spatial resolution is 0.5 mm. Special attention is put on the evaluation of the generalization performance of the investigated models. Results Dose predictions with all models are produced in the order of a second on a GPU, the 3D U‐Net models being fastest with an average of 130 ms. An investigated 3D U‐Net regression model is found to show the strongest performance with overall 61.0%±$\%\pm$0.5% of all voxels exhibiting a deviation in energy deposition prediction of less than 3% compared to full MC simulations with no spatial deviation allowed. The 3D U‐Net models are observed to show better generalization performance for target geometry variations, while the transformer‐based model shows better generalization with regard to the proton energy. Conclusions This paper reveals that (1) all studied deep learning models are significantly faster than non‐machine learning approaches predicting the dose in the order of seconds compared to hours for MC, (2) all models provide reasonable accuracy, and (3) the regression‐trained 3D U‐Net provides the most accurate predictions.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Transformer-based Modelmentioning

confidence: 99%

Section: Transformer-based Modelmentioning

confidence: 99%

Section: Machine Learning Modelsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Small beams, fast predictions: a comparison of machine learning dose prediction models for proton minibeam therapy

et al. 2022

View full text Add to dashboard Cite

show abstract

Sub‐second photon dose prediction via transformer neural networks

et al. 2023

Self Cite

View full text Add to dashboard Cite

Background: Fast dose calculation is critical for online and real-time adaptive therapy workflows. While modern physics-based dose algorithms must compromise accuracy to achieve low computation times, deep learning models can potentially perform dose prediction tasks with both high fidelity and speed. Purpose: We present a deep learning algorithm that, exploiting synergies between transformer and convolutional layers, accurately predicts broad photon beam dose distributions in few milliseconds. Methods: The proposed improved Dose Transformer Algorithm (iDoTA) maps arbitrary patient geometries and beam information (in the form of a 3D projected shape resulting from a simple ray tracing calculation) to their corresponding 3D dose distribution. Treating the 3D CT input and dose output volumes as a sequence of 2D slices along the direction of the photon beam, iDoTA solves the dose prediction task as sequence modeling. The proposed model combines a Transformer backbone routing long-range information between all elements in the sequence, with a series of 3D convolutions extracting local features of the data. We train iDoTA on a dataset of 1700 beam dose distributions, using 11 clinical volumetric modulated arc therapy (VMAT) plans (from prostate, lung, and head and neck cancer patients with 194-354 beams per plan) to assess its accuracy and speed. Results: iDoTA predicts individual photon beams in ≈ 50 ms with a high gamma pass rate of 97.72 ± 1.93% (2 mm, 2%). Furthermore, estimating full VMAT dose distributions in 6-12 s, iDoTA achieves state-of -the-art performance with a 99.51 ± 0.66% (2 mm, 2%) pass rate and an average relative dose error of 0.75 ± 0.36%. Conclusions: Offering the millisecond speed prediction per beam angle needed in online and real-time adaptive treatments, iDoTA represents a new state of the art in data-driven photon dose calculation. The proposed model can massively speed-up current photon workflows, reducing calculation times from few minutes to just a few seconds.

show abstract

Extension of matRad with a modified microdosimetric kinetic model for carbon ion treatment planning: Comparison with Monte Carlo calculation

et al. 2023

View full text Add to dashboard Cite

BackgroundTreatment planning is essential for in silico particle therapy studies. matRad is an open‐source research treatment planning system (TPS) based on the local effect model, which is a type of relative biological effectiveness (RBE) model.PurposeThis study aims to implement a microdosimetric kinetic model (MKM) in matRad and develop an automation algorithm for Monte Carlo (MC) dose recalculation using the TOPAS code. In addition, we provide the developed MKM extension as open‐source tool for users.MethodsCarbon beam data were generated using TOPAS MC pencil beam irradiation. We parameterized the TOPAS MC beam data with a double‐Gaussian fit and modeled the integral depth doses and lateral spot profiles in the range of 100–430 MeV/u. To implement the MKM, the specific energy data table for Z = 1–6 and integrated depth‐specific energy data were acquired based on the Kiefer–Chatterjee track structure and TOPAS MC simulation, respectively. Generic data were integrated into matRad, and treatment planning was performed based on these data. The optimized plan parameters were automatically converted into MC simulation input. Finally, the matRad TPS and TOPAS MC simulations were compared using the RBE‐weighted dose calculation results. A comparison was made for three geometries: homogeneous water phantom, inhomogeneous phantom, and patient.ResultsThe RBE‐weighted dose (DRBE) distribution agreed with TOPAS MC within 1.8% for all target sizes for the homogeneous phantom. For the inhomogeneous phantom, the relative difference in the range of 80% of the prescription dose in the distal fall‐off region (R80) between the matRad TPS and TOPAS MC was 0.6% (1.1 mm). DRBE between the TPS and the MC was within 4.0%. In the patient case, the difference in the dose–volume histogram parameters for the target volume between the TPS and the MC was less than 2.7%. The relative difference in R80 was 0.7% (1.2 mm).ConclusionsThe MKM was successfully implemented in matRad TPS, and the RBE‐weighted dose was comparable to that of TOPAS MC. The MKM‐implemented matRad was released as an open‐source tool. Further investigations with MC simulations can be conducted using this tool, providing a good option for carbon ion research.

show abstract

Millisecond speed deep learning based proton dose calculation with Monte Carlo accuracy

Cited by 27 publications

References 70 publications

Small beams, fast predictions: a comparison of machine learning dose prediction models for proton minibeam therapy

Small beams, fast predictions: a comparison of machine learning dose prediction models for proton minibeam therapy

Sub‐second photon dose prediction via transformer neural networks

Extension of matRad with a modified microdosimetric kinetic model for carbon ion treatment planning: Comparison with Monte Carlo calculation

Contact Info

Product

Resources

About