Incorporating human and learned domain knowledge into training deep neural networks: A differentiable dose volume histogram and adversarial inspired framework for generating Pareto optimal dose distributions in radiation therapy
UT Southwestern Medical Center（西南大学医学中心）
Dan Nguyen, Rafe McBeth, Azar Sadeghnejad Barkousaraie, Gyanendra Bohara, Chenyang Shen, Xun Jia, Steve Jiang
We propose a novel domain specific loss, which is a differentiable loss function based on the dose volume histogram, and combine it with an adversarial loss for the training of deep neural networks to generate Pareto optimal dose distributions. The mean squared error (MSE) loss, dose volume histogram (DVH) loss, and adversarial (ADV) loss were used to train 4 instances of the neural network model: 1) MSE, 2) MSE+ADV, 3) MSE+DVH, and 4) MSE+DVH+ADV. 70 prostate patients were acquired, and the dose influence arrays were calculated for each patient. 1200 Pareto surface plans per patient were generated by pseudo-randomizing the tradeoff weights (84,000 plans total). We divided the data into 54 training, 6 validation, and 10 testing patients. Each model was trained for 100,000 iterations, with a batch size of 2. The prediction time of each model is 0.052 seconds. Quantitatively, the MSE+DVH+ADV model had the lowest prediction error of 0.038 (conformation), 0.026 (homogeneity), 0.298 (R50), 1.65% (D95), 2.14% (D98), 2.43% (D99). The MSE model had the worst prediction error of 0.134 (conformation), 0.041 (homogeneity), 0.520 (R50), 3.91% (D95), 4.33% (D98), 4.60% (D99). For both the mean dose PTV error and the max dose PTV, Body, Bladder and rectum error, the MSE+DVH+ADV outperformed all other models. All model's predictions have an average mean and max dose error less than 2.8% and 4.2%, respectively. Expert human domain specific knowledge can be the largest driver in the performance improvement, and adversarial learning can be used to further capture nuanced features. The real-time prediction capabilities allow for a physician to quickly navigate the tradeoff space, and produce a dose distribution as a tangible endpoint for the dosimetrist to use for planning. This can considerably reduce the treatment planning time, allowing for clinicians to focus their efforts on challenging cases.
University of Texas Southwestern Medical Center（德克萨斯大学西南医学中心）
With the advancement of treatment modalities in radiation therapy for cancer patients, outcomes have improved, but at the cost of increased treatment plan complexity and planning time. The accurate prediction of dose distributions would alleviate this issue by guiding clinical plan optimization to save time and maintain high quality plans. We have modified a convolutional deep network model, U-net (originally designed for segmentation purposes), for predicting dose from patient image contours of the planning target volume (PTV) and organs at risk (OAR). We show that, as an example, we are able…
Dose volume histograms are a useful tool in state-of-the-art radiotherapy planning, and it is essential to be aware of their limitations. Dose distributions computed by treatment planning systems are affected by several sources of uncertainty such as algorithm limitations, measurement uncertainty in the data used to model the beam and residual differences between measured and computed dose, once the model is optimized. In order to take into account the effect of uncertainty, a probabilistic approach is proposed and a new kind of histogram, a dose-expected volume histogram, is introduced. The e…
Therapeutic protons acting on O18-substituted thymidine increase cytotoxicity in radio-resistant human cancer cells. We consider here the physics behind the irradiation during proton beam therapy and diagnosis using O18-enriched thymine in DNA, with attention to the effect of the presence of thymine-18 on cancer cell death.
Purpose: We present a framework for robust automated treatment planning using machine learning, comprising scenario-specific dose prediction and robust dose mimicking. Methods: The scenario dose prediction pipeline is divided into the prediction of nominal dose from input image and the prediction of scenario dose from nominal dose, each using a deep learning model with U-net architecture. By using a specially developed dose-volume histogram-based loss function, the predicted scenario doses are ensured sufficient target coverage despite the possibility of the training data being non-robust. Del…
The secondary neutron fields at the deep tumor therapy terminal at HIRFL (Heavy Ion Research Facility in Lanzhou) were investigated. The distributions of neutron ambient dose equivalent were measured with a FHT762 Wendi-II neutron ambient dose equivalent meter as the 12C ions with energies of 165, 207, 270, and 350 MeV/u bombarded on thick tissue-like targets. The thickness of targets used in experiments is larger than the range of the carbon ions. The neutron spectra and dose equivalent is simulated by using FLUKA code, and the results agree well with the experimental data. The experiment res…
In this research, we presented the method of inhalation therapy with noble gas micro-doses and results of clinical studies in hospital of Russian Academy of Sciences (RAS). We have designed a device that allows dosed injection of noble gas with a volume of about 2-4 ml with peroxide vapors into the nasal cavity. After testing on the authors of the project and obtaining a positive impact of xenon and krypton, a study was conducted on a group of conditionally healthy volunteers in the amount of 20 people. We have noted a positive effect from 18 people. And also, in this article we briefly descri…
MD-Dose: A diffusion model based on the Mamba for radiation dose prediction
Chengdu Institute of Computer Application, Chinese Academy of Sciences（中国科学院成都计算机应用研究所）；School of Computer Science and Technology, University of Chinese Academy of Sciences（中国科学院大学计算机科学与技术学院）；Sichuan University West China Hospital（四川大学华西医院）；Radiophysical Technology Center, Cancer Center, West China Hospital, Sichuan University（四川大学华西医院肿瘤中心放射物理技术中心）
In this work, we propose a Machine Learning model that generates an adjustable 3D dose distribution for external beam radiation therapy for head-and-neck cancer treatments. In contrast to existing Machine Learning methods that provide a single model, we create pairs of models for each organ-at-risk, namely lower-extreme and upper-extreme models. These model pairs for an organ-at-risk propose doses that give lower and higher doses to that organ-at-risk, while also encapsulating the dose trade-off to other organs-at-risk. By weighting and combining the model pairs for all organs-at-risk, we are …
We present a method of directly optimizing on deviations in clinical goal values in radiation therapy treatment planning. Using a new mathematical framework in which metrics derived from the dose-volume histogram are regarded as functionals of an auxiliary random variable, we are able to obtain volume-at-dose and dose-at-volume as infinitely differentiable functions of the dose distribution with easily evaluable function values and gradients. Motivated by the connection to risk measures in finance, which is formalized in this framework, we also derive closed-form formulas for mean-tail-dose an…
UT Southwestern Medical Center（德克萨斯大学西南医学中心）；University of North Texas（北德克萨斯大学）；UCLouvain（鲁汶大学）；Erasmus University Medical Center（伊拉斯姆斯大学医学中心）
This work aims to study the generalizability of a pre-developed deep learning (DL) dose prediction model for volumetric modulated arc therapy (VMAT) for prostate cancer and to adapt the model to three different internal treatment planning styles and one external institution planning style. We built the source model with planning data from 108 patients previously treated with VMAT for prostate cancer. For the transfer learning, we selected patient cases planned with three different styles from the same institution and one style from a different institution to adapt the source model to four targ…
A new strategy for radiation therapy dosimetry planning (RTDP) used to reduce dose estimation errors due to respiratory motion in breast treatment was illustrated and evaluated in this study. On CT data set acquired for breast treatment, six different RTDP tangential techniques were performed: (i) three-dimensional conformal radiotherapy (3DCRT) with physical wedge, (ii) 3DCRT with virtual wedge, (iii) motion management technique (MMT) with physical wedge, (iv) MMT with virtual wedge, (v) 3DCRT with field-in-field, and (vi) intensity-modulated radiation therapy (IMRT) with direct aperture opti…
To accurately verify the dose of intensity-modulated radiation therapy (IMRT), we have used a global optimization method to investigate a new dose-verification algorithm. In practical application of this quality assurance (QA) procedure, verification of the dose using calculated and measured dose distributions involves a subtle problem in the region of high dose gradient. Consideration of systematic errors shows that the large dose differences in high-dose-gradient regions are due to the unexpected shift of measuring devices. We have proposed an optimization algorithm to correct this error, an…
The neutron dose distributions on observation distances and on observation angles were measured using a Wendi-II neutron dose-meter at the deep tumor therapy terminal at HIRFL (Heavy Ion Research Facility in Lanzhou) as 12C ions with energies from 165 to 350 MeV/u bombarding on thick solid water targets with different thickness according to the ion energies. The experimental results were compared with those calculated by FLUKA code. It is found that the experimental data was in good agreement with the calculated results. The neutron energy spectra were also studied by using the FLUKA code. The…
Secondary radiation in a radiation therapy environment is considered a cause for post-treatment side effects including secondary carcinogenesis. In a proton-therapy setup, neutrons deliver a considerable extra dose to the patient. Various measurements and calculations of neutron dose suggest a wide range of values for this quantity. This paper provides a comparison between the previously published values and analyzes the Monte Carlo calculated neutron dose from the present work using the neutron cross-section data. The purpose is to provide a range for the convergence of several neutron-absorb…
Fast and accurate dose predictions are one of the bottlenecks in treatment planning for microbeam radiation therapy (MRT). In this paper, we propose a machine learning (ML) model based on a 3D U-Net. Our approach predicts separately the large doses of the narrow high intensity synchrotron microbeams and the lower valley doses between them. For this purpose, a concept of macro peak doses and macro valley doses is introduced, describing the respective doses not on a microscopic level but as macroscopic quantities in larger voxels. The ML model is trained to mimic full Monte Carlo (MC) data. Comp…
Purpose: The presence of respiratory motion during radiation treatment leads to degradation of the expected dose distribution, both for target coverage and healthy-tissue sparing, particularly for techniques like pencil-beam scanning proton therapy which have dynamic delivery systems. While tools exist to estimate this degraded four-dimensional (4D) dose, they typically have one or more deficiencies such as ... Methods: To quickly compute the 4D-dose, the three main tasks of the calculator were run on graphics processing units (GPUs). These tasks were: simulating the delivery of the plan using…
Mayo Clinic（梅奥诊所）；University of Georgia（佐治亚大学）；Technische Universität Dresden（德累斯顿工业大学）；Helmholtz-Zentrum Dresden - Rossendorf（德累斯顿-罗森多夫亥姆霍兹中心）；University of Virginia（弗吉尼亚大学）
Purpose: We propose a general framework for quantifying predictive uncertainties of dose-related quantities and leveraging this information in a dose mimicking problem in the context of automated radiation therapy treatment planning. Methods: A three-step pipeline, comprising feature extraction, dose statistic prediction and dose mimicking, is employed. In particular, the features are produced by a convolutional variational autoencoder and used as inputs in a previously developed nonparametric Bayesian statistical method, estimating the multivariate predictive distribution of a collection of p…
Cone beam CT (CBCT) has been widely used for patient setup in image guided radiation therapy (IGRT). Radiation dose from CBCT scans has become a clinical concern. The purposes of this study are 1) to commission a GPU-based Monte Carlo (MC) dose calculation package gCTD for Varian On-Board Imaging (OBI) system and test the calculation accuracy, and 2) to quantitatively evaluate CBCT dose from the OBI system in typical IGRT scan protocols. We first conducted dose measurements in a water phantom. X-ray source model parameters used in gCTD are obtained through a commissioning process. gCTD accurac…
Radiotherapy of pregnant cancer patients is not common, but when applied accurate assessment of the fetus dose is required, especially since the treatment planning systems are not able and do not allow accurate assessment of out-of-field doses. Proton therapy significantly reduces out-of-field doses, such as dose to the fetus, when compared to the photon radiotherapy techniques and as such could be promising for pregnant cancer patients. Within this study Monte Carlo calculations are performed on the three different computational phantoms representing pregnant women, all in second trimester of…
We demonstrate the application of mixture density networks (MDNs) in the context of automated radiation therapy treatment planning. It is shown that an MDN can produce good predictions of dose distributions as well as reflect uncertain decision making associated with inherently conflicting clinical tradeoffs, in contrast to deterministic methods previously investigated in literature. A two-component Gaussian MDN is trained on a set of treatment plans for postoperative prostate patients with varying extents to which rectum dose sparing was prioritized over target coverage. Examination on a test…
Conventional cancer therapies include surgery, radiation therapy, chemotherapy, and, more recently, immunotherapy. These modalities are often combined to improve the therapeutic index. The general concept of radiation therapy is to increase the therapeutic index by creating a physical dose differential between tumors and normal tissues through precision dose targeting, image guidance, and high radiation beams that deliver radiation dose with high conformality, e.g., protons and ions. However, treatment and cure are still limited by normal tissue radiation toxicity, with many patients experienc…
Charged Particle Therapy is a technique for cancer treatment that exploits hadron beams, mostly protons and carbons. A critical issue is the monitoring of the dose released by the beam to the tumor and to the surrounding tissues. We present the design of a new tracking device for monitoring on-line the dose in ion therapy through the detection of secondary charged particles produced by the beam interactions in the patient tissues. In fact, the charged particle emission shape can be correlated with the spatial dose release and the Bragg peak position. The detector uses the information provided …
For the in vivo range verification in proton therapy, it has been tried to measure the spatial distribution of the prompt gammas generated by the proton-induced interactions with the close relationship with the proton dose distribution. However, the high energy of the prompt gammas and background gammas are still problematic in measuring the distribution. In this study, we suggested a new method determining the in vivo range by utilizing the time structure of the prompt gammas formed with the rotation of a range modulation wheel (RMW) in the passive scattering proton therapy. To validate the M…
A split feasibility formulation for the inverse problem of intensity-modulated radiation therapy (IMRT) treatment planning with dose-volume constraints (DVCs) included in the planning algorithm is presented. It involves a new type of sparsity constraint that enables the inclusion of a percentage-violation constraint in the model problem and its handling by continuous (as opposed to integer) methods. We propose an iterative algorithmic framework for solving such a problem by applying the feasibility-seeking CQ-algorithm of Byrne combined with the automatic relaxation method (ARM) that uses cycl…
Purpose: To verify dose delivery and quality assurance of volumetric modulated arc therapy (VMAT) for head and neck cancer. Method: The Imaging and Radiation Oncology Core Houston (IROC-H) head and neck phantom with thermo- luminescent dosimeters (TLDs) and films, were imaged with computed tomography scan and the reconstructed image was transferred to pinnacle treatment planning system (TPS). On TPS the planning target volume (PTV), secondary target volume (STV) and organ at risk (OAR) were delineated manually and a treatment plan was made. The dose constraints were determined for the concerne…
Cao et al (2018) published an article on inverse planning based on dose-averaged linear energy transfer (LET). Their claim that the problem can be cast as a linear optimization model relies on an incorrect application of the Charnes-Cooper transformation. In this comment we show that their linear model is simlar to one where dose-averaged LET is multiplied with dose, explaining why their model was nonetheless able to improve the LET distribution.
University of Virginia Health System（弗吉尼亚大学健康系统）；Thomas Jefferson University Hospital（托马斯杰斐逊大学医院）
The DeepDoseNet 3D dose prediction model based on ResNet and Dilated DenseNet is proposed. The 340 head-and-neck datasets from the 2020 AAPM OpenKBP challenge were utilized, with 200 for training, 40 for validation, and 100 for testing. Structures include 56Gy, 63Gy, 70Gy PTVs, and brainstem, spinal cord, right parotid, left parotid, larynx, esophagus, and mandible OARs. Mean squared error (MSE) loss, mean absolute error (MAE) loss, and MAE plus dose-volume histogram (DVH) based loss functions were investigated. Each model's performance was compared using a 3D dose score, $\bar{S_{D}}$, (mean …
We give a scenario-based treatment plan optimization formulation that is equivalent to planning with geometric margins if the scenario doses are calculated using the static dose cloud approximation. If the scenario doses are instead calculated more accurately, then our formulation provides a novel robust planning method that overcomes many of the difficulties associated with previous scenario-based robust planning methods. In particular, our method protects only against uncertainties that can occur in practice, it gives a sharp dose fall-off outside high dose regions, and it avoids underdosage…
We study the propagation of nucleons and nuclei in tissue-like media within a Monte Carlo Model for Heavy-ion Therapy (MCHIT) based on the GEANT4 toolkit (version 8.2). The model takes into account fragmentation of projectile nuclei and secondary interactions of produced nuclear fragments. Model predictions are validated with available experimental data obtained for water and PMMA phantoms irradiated by monoenergetic carbon-ion beams. The MCHIT model describes well (1) the depth-dose distributions in water and PMMA, (2) the doses measured for fragments of certain charge, (3) the distributions …
Beam modelling of Hitachi PROBEAT proton therapy system for a GPU-based Fast Monte Carlo dose engine
University of Nebraska Omaha（内布拉斯加大学奥马哈分校）；Rice University（莱斯大学）；The University of Texas MD Anderson Cancer（德克萨斯大学安德森癌症中心）；Mevion Medical System（美维医疗系统）；Emory University（埃默里大学）
Recently, artificial intelligence technologies and algorithms have become a major focus for advancements in treatment planning for radiation therapy. As these are starting to become incorporated into the clinical workflow, a major concern from clinicians is not whether the model is accurate, but whether the model can express to a human operator when it does not know if its answer is correct. We propose to use Monte Carlo dropout (MCDO) and the bootstrap aggregation (bagging) technique on deep learning models to produce uncertainty estimations for radiation therapy dose prediction. We show that…
We propose to develop deep learning models that can predict Pareto optimal dose distributions by using any given set of beam angles, along with patient anatomy, as input to train the deep neural networks. We implement and compare two deep learning networks that predict with two different beam configuration modalities. We generated Pareto optimal plans for 70 patients with prostate cancer. We used fluence map optimization to generate 500 IMRT plans that sampled the Pareto surface for each patient, for a total of 35,000 plans. We studied and compared two different models, Model I and Model II. M…
Carbon ions offer significant advantages for deep-seated local tumors therapy due to their physical and biological properties. Secondary particles, especially neutrons caused by heavy ion reactions should be carefully considered in treatment process and radiation protection. For radiation protection purposes, the FLUKA Code was used in order to evaluate the radiation field at deep tumor therapy room of HIRFL in this paper. The neutron energy spectra, neutron dose and energy deposition of carbon ion and neutron in tissue-like media was studied for bombardment of solid water target by 430MeV/u C…
Two full rotating gantry with different nozzles (Multipurpose nozzle with MLC, Scanning Dedicated nozzle) with conventional cyclotron system is installed and under commissioning for various proton treatment options at Samsung Medical Center in Korea. The purpose of this study is to investigate neutron dose equivalent per therapeutic dose, H/D, to x-ray imaging equipment under various treatment conditions with monte carlo simulation. At first, we investigated H/D with the various modifications of the beam line devices (Scattering, Scanning, Multi-leaf collimator, Aperture, Compensator) at isoce…
In this study, two proton beam delivery designs, passive scattering proton therapy (PSPT) and pencil beam scanning (PBS), were quantitatively compared in terms of dosimetric indices. The GATE Monte Carlo code was used to simulate the proton beam system; and the developed simulation engines were benchmarked with respect to the experimental measurements. A water phantom was used to simulate system energy parameters using a set of depth-dose data in the energy range of 120-235 MeV. To compare the performance of PSPT against PBS, multiple dosimetric parameters including FWHM, peak position, range,…
In intensity-modulated radiation therapy, optimal intensity distributions of incoming beams are decomposed into linear combinations of leaf openings of a multileaf collimator (segments). In order to avoid inefficient dose delivery, the decomposition should satisfy a number of dosimetric constraints due to suboptimal dose characteristics of small segments. However, exact decomposition with dosimetric constraints is only in limited cases possible. The present work introduces new heuristic segmentation algorithms for the following optimization problem: Find a segmentation of an approximated matri…
We compared the treatment planning performance of RapidArc (RA) vs. CyberKnife (CK) for spinal stereotactic body radiation therapy (SBRT). Ten patients with spinal lesions who had been treated with CK were re-planned with RA, which consisted of two complete arcs. Computed tomography (CT) and volumetric dose data of CK, generated using the Multiplan (Accuray) treatment planning system (TPS) and the Ray-Trace algorithm, were imported to Varian Eclipse TPS in Dicom format, and the data were compared with the RA plan using analytical anisotropic algorithm (AAA) dose calculation. The optimized dose…
Carbon ion therapy have the ability to overcome the limitation of convertional radiotherapy due to its most energy deposition in selective depth, usually called Bragg peak, which results in increased biological effectiness. During carbon ion therapy, lots positron emitters such as $^{11}$C, $^{15}$O, $^{10}$C are generated in irradiated tissues by nuclear reactions. Immediately after patient irradiation, PET scanners can be used to measure the spatial distribution of positron emitters, which can track the carbon beam to the tissue. In this study, we designed and evaluated an dual-plate in-room…
