Purpose: The purpose of this study is to investigate the effect of different magnetic resonance (MR) sequences on the accuracy of deep learning-based synthetic computed tomography (sCT) generation in the complex head and neck region. Methods: Four sequences of MR images (T1, T2, T1C, and T1DixonC-water) were collected from 45 patients with nasopharyngeal carcinoma. Seven conditional generative adversarial network (cGAN) models were trained with different sequences (single channel) and different combinations (multi-channel) as inputs. To further verify the cGAN performance, we also used a U-net network as a comparison. Mean absolute error, structural similarity index, peak signal-to-noise ratio, dice similarity coefficient, and dose distribution were evaluated between the actual CTs and sCTs generated from different models. Results: The results show that the cGAN model with multi-channel (i.e., T1 + T2 + T1C + T1DixonC-water) as input to predict sCT achieves higher accuracy than any single MR sequence model. The T1-weighted MR model achieves better results than T2, T1C, and T1DixonC-water models. The comparison between cGAN and U-net shows that the sCTs predicted by cGAN retains additional image details are less blurred and more similar to the actual CT. Conclusions: Conditional generative adversarial network with multiple MR sequences as model input shows the best accuracy. The T1-weighted MR images provide sufficient image information and are suitable for sCT prediction in clinical scenarios with limited acquisition sequences or limited acquisition time.

No abstract

Purpose: To investigate region-specific models for organ's three-dimensional dose distribution prediction with neural network. Methods: The dose distribution from different bladder regions for 52 prostate volumetric modulated arc therapy cases were first analyzed, the two region-specific models were then built to predict the bladder dose distribution, the initial model and the refined model. For the initial model, the bladder was divided into overlapping region and nonoverlapping region, two artificial neural networks were established with each one corresponding to one region. For the refined model, the nonoverlapping region was further divided into three subregions, and four artificial neural network models were built in total. For each artificial neural network model, several spatial and volumetric features for the bladder were extracted as the input to the neural network. To investigate the feasibility and dose distribution prediction accuracy of the proposed two region-specific models, the mean absolute error, gamma passing rate, dose volume histogram, and dose distribution for the refined model were compared. Results: According to the predicted dose from the initial model and the refined model, the average mean absolute error for all cases is reduced from 5.03 Gy in the initial model to 3.23 Gy in the refined model, the refined model reduce the mean absolute error about 2% relative to the prescription dose. The average area deviation of predicted dose volume histogram by the refined model is 5%, and the average gamma passing rate is 82% and 94% with the 3 mm/3% and 5 mm/5% criteria, which shows that the refined model proposed in this study has high dose-prediction accuracy. Conclusions: Two region-specific three-dimensional dose distribution prediction models for volumetric modulated arc therapy prostate cases based on neural network have been investigated, the models have shown that a more refined consideration of structures improved the accuracy of predicted dose distribution.

Mengke Qi and Yongbao Li equally contributed to the work as co-first authors.

In current knowledge-based treatment planning for intensity-modulated radiation therapy (IMRT), 3-dimensional dosimetric goals are predicted to provide abundant and appropriate starting points for planning optimization, but considering there're uncertainties with those dose distribution predictions, how to tailor the objective function and constraints accordingly is quite a concern. Here, we represent a novel automatic treatment optimization method that is capable of making the most of dose distribution prediction meanwhile achieving its optimum as much as possible. On the foundation of an in-house organsat-risk (OARs) dose distribution prediction model, we reformulate a traditional fluence map optimization (FMO) model by a predicted dose distribution-based objective, an equivalent uniform dose sparing for OARs and hard dose constraints for planning target volume (PTV). Feasibility and performance of the method is evaluated with 10 gynecology (GYN) cancer IMRT cases by comparing the plan quality of the generated to the original clinical ones, in the term of dose-volume-histogram (DVH) curves, dose distribution and detailed dosimetric endpoints. Results show plan quality improvement by our proposed method, with comparable PTV dose coverage but further dose sparing for OARs. Among 6 investigated OAR dosimetric endpoints, 4 of them are observed with significant improvement (P<0.05), V 30 , V 45 of rectum is decreased by (8.42±7.88) %, (15.49±7.48) %, respectively and V 30 , V 45 of bladder is decreased by (14.47±5.08) %, (14.24±4.71) %, respectively. We have successfully developed a novel automatic optimization method which is able to make good use of 3D dose prediction and ensure the output plan quality for IMRT. INDEX TERMS 3D dose distribution prediction, equivalent uniform dose, intensity modulated radiation therapy, prediction guided treatment planning optimization.

Song (2020) Robustness comparative study of dose-volume-histogram prediction models for knowledge-based radiotherapy treatment planning, ABSTRACT Purpose: To compare the robustness of three dose-volume-histogram (DVH) prediction models for knowledge-based treatment planning (KBP) for radiation therapy. Methods: Three models proposed by Zhu et al. (Zmodel), Lindsey et al. (Lmodel), and Satomi et al. (Smodel) were selected, and compared based on identified archived radiation therapy plan cohorts (including 50 prostate cancer (PC) and 29 nasopharynx cancer (NPC) cases). Robustness comparison was performed by observing changes in prediction accuracy in relation to training example size, and further analyzing the number of training samples required for each model. In addition, a robustness comparison of models on different case applications was conducted to verify the applicability of models on different tumor sites. The error of model predictions was measured by the difference between predicted and clinical DVH. Results: The minimum necessary datasets required to train the model are 35, 40, and 45 for Lmodel, Zmodel, and Smodel, respectively. Smodel has high accuracy on both PC and NPC databases, achieving a median prediction error of 0.0257 on the training dataset and 0.0446 on the evaluation dataset. In a specific case, Smodel and Zmodel exhibit the best result on PC (with prediction errors of 0.0464) and NPC case applications (with prediction errors of 0.0228), respectively. Conclusions: Lmodel needs the least number of samples necessary for training. Smodel and Zmodel are optimal for the PC and NPC cases, respectively. In different case applications, Smodel performs more stable. Planners or researchers should carefully select an appropriate method under specific requirements. ARTICLE HISTORY

Objective. To develop a novel deep learning-based 3D in vivo dose reconstruction framework with an electronic portal imaging device (EPID) for magnetic resonance-linear accelerators (MR-LINACs). Approach. The proposed method directly back-projected 2D portal dose into 3D patient coarse dose, which bypassed the complicated patient-to-EPID scatter estimation step used in conventional methods. A pre-trained convolutional neural network (CNN) was then employed to map the coarse dose to the final accurate dose. The electron return effect caused by the magnetic field was captured with the CNN model. Patient dose and portal dose datasets were synchronously generated with Monte Carlo simulation for 96 patients (78 cases for training and validation and 18 cases for testing) treated with fixed-beam intensity-modulated radiotherapy in four different tumor sites, including the brain, nasopharynx, lung, and rectum. Beam angles from the training dataset were further rotated 2–3 times, and doses were recalculated to augment the datasets. Results. The comparison between reconstructed doses and MC ground truth doses showed mean absolute errors <0.88% for all tumor sites. The averaged 3D γ-passing rates (3%, 2 mm) were 97.42%±2.66% (brain), 98.53%±0.95% (nasopharynx), 99.41%±0.46% (lung), and 98.63%±1.01% (rectum). The dose volume histograms and indices also showed good consistency. The average dose reconstruction time, including back projection and CNN dose mapping, was less than 3 s for each individual beam. Significance. The proposed method can be potentially used for accurate and fast 3D dosimetric verification for online adaptive radiotherapy using MR-LINACs.

BackgroundScattering photons can seriously contaminate cone‐beam CT (CBCT) image quality with severe artifacts and substantial degradation of CT value accuracy, which is a major concern limiting the widespread application of CBCT in the medical field. The scatter kernel deconvolution (SKD) method commonly used in clinic requires a Monte Carlo (MC) simulation to determine numerous quality‐related kernel parameters, and it cannot realize intelligent scatter kernel parameter optimization, causing limited accuracy of scatter estimation.PurposeAiming at improving the scatter estimation accuracy of the SKD algorithm, an intelligent scatter correction framework integrating the SKD with deep reinforcement learning (DRL) scheme is proposed.MethodsOur method firstly builds a scatter kernel model to iteratively convolve with raw projections, and then the deep Q‐network of the DRL scheme is introduced to intelligently interact with the scatter kernel to achieve a projection adaptive parameter optimization. The potential of the proposed framework is demonstrated on CBCT head and pelvis simulation data and experimental CBCT measurement data. Furthermore, we have implemented the U‐net based scatter estimation approach for comparison.ResultsThe simulation study demonstrates that the mean absolute percentage error (MAPE) of the proposed method is less than 9.72% and the peak signal‐to‐noise ratio (PSNR) is higher than 23.90 dB, while for the conventional SKD algorithm, the minimum MAPE is 17.92% and the maximum PSNR is 19.32 dB. In the measurement study, we adopt a hardware‐based beam stop array algorithm to obtain the scatter‐free projections as a comparison baseline, and our method can achieve superior performance with MAPE < 17.79% and PSNR > 16.34 dB.ConclusionsIn this paper, we propose an intelligent scatter correction framework that integrates the physical scatter kernel model with DRL algorithm, which has the potential to improve the accuracy of the clinical scatter correction method to obtain better CBCT imaging quality.