The objective of this study was to develop an automatic two-stage diagnostic system for detecting and classifying cervical spine fractures using 3D segmentation and 2.5D classification approaches on CT images. The first stage aimed to achieve accurate 3D segmentation of cervical spine structures, while the second stage focused on classifying fractures at both the individual cervical vertebra level and the overall patient level. The study utilized a dataset of 87 cases with pixel-level semantic segmentation annotations and 2019 CT scans. A 3D segmentation model was employed for automatic segmentation, followed by a fracture classification system using 2.5D deep learning models. Data augmentation and preprocessing techniques were applied to enhance model performance. The models were trained and evaluated using 5-fold cross-validation, with performance metrics including Dice coefficient, Recall, Specificity, F1 Score, AUC, and Accuracy. In the first stage, the 3D segmentation model achieved a Dice coefficient of 0.93 on the test set, indicating high segmentation accuracy for cervical spine structures. In the second stage, the classification models achieved accuracies ranging from 0.8980 to 0.9812 at both the C1–C7 vertebra level and the overall patient level. However, there were noticeable variations in sensitivity and positive predictive value across different vertebral levels, with relatively lower sensitivity observed at C3 and C5–C6. Overall, the proposed two-stage system demonstrated good performance in fracture detection and localization and outperformed previously reported methods on several evaluation metrics. This study developed a two-stage automatic diagnostic system based on 3D segmentation and 2.5D classification for the detection and localization of cervical spine fractures. The system achieved improved diagnostic accuracy and efficiency in our study cohort and showed certain potential for clinical application. Nevertheless, we observed substantial variability in sensitivity and precision across different vertebrae, and the detection performance for some levels (such as C3 and C5–C6) remained suboptimal. In addition, the absence of external validation limits the direct generalizability of the model to clinical practice. Future work should focus on further optimization and validation of the system in larger, more diverse, and multicenter cohorts to enhance its generalizability and clinical utility.

This study aims to develop a mamba-based UNet model (MambaUNet) for fast and accurate volumetric modulated arc therapy (VMAT) plan dose calculation in cervical cancer. The 2-D fluence maps from all control points were projected into a 3-D fluence volume and combined with the corresponding density volume as input to MambaUNet, which subsequently calculated the 3-D dose distribution. The model integrated volumetric feature extraction via a 3-D residual UNet and efficient global dependency modeling through a mamba encoder. A total of 633 cervical cancer VMAT plans were used, with 445 allocated for training, 128 for validation, and 60 for testing. The MambaUNet model was also comparatively evaluated against three other models (iDoTA, TransUNet, and ResUNet) in terms of dose calculation accuracy and computational efficiency. On the test set, the dose-volume histograms (DVHs), mean absolute errors (MAEs), root-mean-square errors (RMSEs), Dice similarity coefficients (DSCs), and dosimetric indices between the model-calculated doses and the ground-truth doses all demonstrated good agreement. The proposed MambaUNet model achieved an average 3-D gamma passing rate of $98.79 similar to\pm similar to 0.78$ % under the 3%/2 mm criterion. It provided dose calculation accuracy comparable to or slightly better than the other models and achieved higher computational efficiency. The entire dose calculation process, including fluence map generation from MLC sequences, fluence projection, and MambaUNet model inference, can be completed within 80 ms on average. These results suggest that MambaUNet has strong potential for accelerating the VMAT planning process.

Nasopharyngeal carcinoma is a prevalent malignancy in the head and neck region, where accurate delineation of the gross tumor volume of the nasopharyngeal lesion and clinical target volume is essential for precise radiotherapy planning. However, existing segmentation approaches often rely on multiple imaging modalities, substantially increasing the clinical burden of image acquisition and manual annotation. This study aims to achieve multimodal-level segmentation performance using only single-modality input, thereby reducing clinical costs while maintaining high accuracy. To this end, we propose a novel two-stage deep learning framework that integrates multimodal synthetic image generation and feature aggregation for cost-effective nasopharyngeal carcinoma tumor segmentation. In the first stage, a re-implemented synchronous generation network synthesizes intermediate multimodal magnetic resonance images from a single T1-weighted scan, effectively reconstructing cross-modality information without requiring additional imaging sequences. In the second stage, a Convolutional neural network-Transformer hybrid network incorporates cross-modality attention and feature aggregation mechanisms to leverage the generative priors for improved tumor boundary representation and contextual understanding. Extensive experiments were conducted on 167 retrospectively collected clinical nasopharyngeal carcinoma cases. The proposed framework achieves mean dice similarity coefficients of 0.81 +/- 0.03 for the gross tumor volume of the nasopharyngeal lesion and 0.85 +/- 0.04 for clinical target volume, comparable to multimodal segmentation networks while requiring only single-modality input. Moreover, the method demonstrates superior robustness and consistency across different tumor stages. This novel framework significantly enhances segmentation accuracy while reducing the dependency on multiple medical imaging modalities, paving the way for low-cost and high-accuracy nasopharyngeal carcinoma diagnosis and treatment.

OBJECTIVES:To obtain high-quality pre-treatment localization MR (sMR) images from dynamic cine-MR using the Swin-ResViT network for target tracking in MRgRT. METHODS:We propose a ResViT model fused with a Swin Transformer module (Swin-ResViT) with an optimized bottleneck layer structure for enhancing feature extraction efficiency. Seventeen liver cancer patients were retrospectively enrolled from Sun Yat-sen University Cancer Center from February to July 2024, and 12 of them were assigned to the training set (using intra-treatment cine-MR and pre-treatment planning MR), with the remaining 5 patients as the test set. Image generation quality and model performance were comprehensively evaluated by quantifying the normalized root mean square error (NRMSE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), motion marker point error, and model inference speed between sMR and reference localization MR. RESULTS:Regarding image quality, Swin-ResViT reduced NRMSE and LPIPS by 90% and 82% compared to cine-MR (P<0.001), and improved PSNR, SSIM, and CNR by 157%, 79%, and 181% (P<0.001), respectively. Regarding structural accuracy, the mean localization error of motion markers at the right hepatophrenic junction in the generated dynamic sMR sequences was 0.7695±0.7294 mm (P<0.05). Regarding model inference speed, for a single 224×224-pixel frame, the average processing time on an NVIDIA GeForce RTX 2080 Ti GPU was 15.5 ms for Swin-ResViT as compared with 41.4 ms for the ResViT network, demonstrating a 62% reduction. CONCLUSIONS:The Swin-ResViT model can synthesize high-quality sMR from cine-MR images. This method combines computational efficiency with significant image enhancement advantages, and thus has important clinical significance for real-time MRgRT.

Radiation-induced cardiac toxicity remains a significant concern in breast cancer radiotherapy. Our institution developed a novel lateral decubitus position (nLDP) technique for patients unable to maintain deep inspiration breath-hold (DIBH) during postoperative radiotherapy (PORT). This study quantitatively evaluated the cardiac dosimetry benefits and setup errors of nLDP and compared its heart-sparing effects to DIBH within the low-risk patients (free breathing mean heart dose [FB–MHD] <4 Gy). Patients with left-sided breast cancer indicated for PORT were enrolled and stratified into three cohorts based on positioning and breathing techniques: Cohort 1 (nLDP in FB, nLDP-FB), Cohort 2 (supine position in FB, SP-FB), and Cohort 3 (SP in DIBH, SP-DIBH). The planning target volume (PTV) encompassed the whole breast/chest wall ± regional lymph nodes, and organs at risk (OARs) were delineated according to institutional protocols. While all patients were clinically treated without internal mammary node irradiation (IMNI), supplementary IMN-inclusive plans were generated specifically for Cohort 1 for dosimetric comparison. All plans utilized 6 MV photon intensity-modulated radiation therapy (IMRT). Dosimetric parameters, normal tissue complication probability (NTCP), anatomical heart-PTV distances, and setup errors were assessed. For the low-risk patients, cardiac dose reductions in nLDP were compared with DIBH. A total of 76 patients were analyzed: Cohort 1 (n = 28, median age 51.5 years, 75

PURPOSE:Accurate prediction of distant metastasis (DM) and locoregional recurrence (LR) in head and neck cancer (HNC) patients following radiotherapy is critical for individualized treatment planning. This study aimed to develop and validate an explainable deep learning model integrating anatomical and dosimetric information for HNC DM and LR risk stratification. METHODS AND MATERIALS:237 HNC patients treated with definitive radiotherapy were collected and divided into training, internal, and external validation cohorts. A dual-task deep learning model based on 3D squeeze-and-excitation residual networks was constructed to predict DM and LR risk using CT images, 3D dose distributions, and gross tumor volume (GTV) masks. Model performance was evaluated by concordance index (C-index), time-dependent ROC, and decision curve analysis. Interpretability was enhanced using Grad-CAM and a novel Activation Volume Histogram (AVH) method to quantify attention patterns across spatial-dosimetric subregions. RESULTS:The model achieved C-indices of 0.92 and 0.74 for DM and LR, respectively, outperforming radiomic models. Grad-CAM visualizations revealed distinct activation patterns aligned with recurrence sites- lymphatic regions for DM and tumor zones for LR. AVH analysis identified GTV extended by 3mm receiving dose ≥ 65 Gy and GTV receiving dose ≥ 65 Gy as the most discriminative subregions, showing significant differences in activation distributions between high- and low-risk groups, particularly within the 0.4-0.5 intensity range. CONCLUSION:This dual-task model enables accurate DM risk prediction and showed potential for LR risk prediction in HNC, providing spatial-dosimetric information to support risk-adaptive radiotherapy. Further extensive multicenter validation is warranted to confirm its clinical applicability.

PURPOSE:Monte Carlo (MC)-based dose calculation provides high accuracy in radiotherapy but is limited by statistical uncertainty (SU), which introduces noise and increases computational demands. This study proposes a Statistical Uncertainty-aware deep learning framework with a Dual-Path Dilated Convolution Fusion architecture to improve the accuracy and efficiency of MC dose denoising. METHODS:We designed a three-channel convolutional neural network that integrates noisy MC dose distributions, corresponding SU maps, and CT images. The dual-path structure combines standard and dilated convolutions to extract both local anatomical features and global contextual information. The model was trained and validated on 69 clinical IMRT plans from three tumor sites (head-and-neck, brain, and lung), each with six levels of simulated noise generated by a GPU-based MC engine. High-particle MC doses served as the ground truth. A total of six sub-models were trained for different noise levels, and performance was evaluated using mean dose error (MDE), gamma passing rate (GPR), and dose-volume histogram (DVH) analysis. RESULTS:The SU-aware model outperformed its two-channel SU-agnostic counterpart across all tumor sites, with MDEs reduced to (0.84 ± 0.26)% for head-and-neck, (0.85 ± 0.21)% for brain, and (0.36 ± 0.09)% for lung. GPRs exceeded 97 % (3 %/3 mm) with ≥ 1 × 105 histories. Denoising was completed within seconds, enabling real-time application. CONCLUSIONS:By explicitly incorporating SU into the network, the proposed model achieves fast, accurate dose denoising.

Objective: The objective of this study was to develop and assess the clinical feasibility of auto-segmentation and auto-planning methodologies for automated radiotherapy in prostate cancer. Methods: A total of 166 patients were used to train a 3D Unet model for segmentation of the gross tumor volume (GTV), clinical tumor volume (CTV), nodal CTV (CTVnd), and organs at risk (OARs). Performance was assessed by the Dice similarity coefficient (DSC), the Recall, Precision, Volume Ratio (VR), the 95% Hausdorff distance (HD95%), and the volumetric revision degree (VRD). An auto-planning network based on a 3D Unet was trained on 77 treatment plans derived from the 166 patients. Dosimetric differences and clinical acceptability of the auto-plans were studied. The effect of OAR editing on dosimetry was also evaluated. Results: On an independent set of 50 cases, the auto-segmentation process took 1 min 20 s per case. The DSCs for GTV, CTV, and CTVnd were 0.87, 0.88, and 0.82, respectively, with VRDs ranging from 0.09 to 0.14. The segmentation of OARs demonstrated high accuracy (DSC ≥ 0.83, Recall/Precision ≈ 1.0). The auto-planning process required 1–3 optimization iterations for 50%, 40%, and 10% of cases, respectively, and exhibited significant better conformity (p ≤ 0.01) and OAR sparing (p ≤ 0.03) while maintaining comparable target coverage. Only 6.7% of auto-plans were deemed unacceptable compared to 20% of manual plans, with 75% of auto-plans considered superior. Notably, the editing of OARs had no significant impact on doses. Conclusions: The accuracy of auto-segmentation is comparable to that of manual segmentation, and the auto-planning offers equivalent or better OAR protection, meeting the requirements of online automated radiotherapy and facilitating its clinical application.

OBJECTIVES:To quantitatively analyze setup errors of 4 immobilization devices in precision radiotherapy for prostate cancer, their accuracy differences, and the factors affecting their setup precisions. METHODS:We conducted a retrospective analysis of 240 prostate cancer patients undergoing image-guided radiotherapy at Sun Yat-sen University Cancer Center from May, 2016 to May, 2024. According to the immobilization devices used, the patients were divided into 1.2 m vacuum bag group (n=60), 1.8 m vacuum bag group (n=60), Orfit frame group (n=60), and customized prone board group (n=60). All the patients received pre-treatment cone-beam CT (CBCT) scans, and setup errors in the right-left (RL), superior-inferior (SI), and anterior-posterior (AP) directions were obtained through XVI system grayscale registration. Further subgroup analyses were performed based on patient stratifications by lymph node irradiation status (n=120 each), age (<65 years, n=80; ≥65 years, n=160), and BMI (BMI<24 kg/m², n=120; BMI≥24 kg/m², n=120). RESULTS:The setup errors differed significantly among the 4 groups in three-dimensional directions (P<0.05). The customized prone board group showed minimal errors in the RL (0.02±0.25 cm) and SI (0.01±0.32 cm) directions, but demonstrated the largest error in the AP direction (-0.28±0.36 cm). The patients with lymph node irradiation had significantly greater AP directional errors (-0.22±0.36 cm) than those without (-0.01±0.43 cm; P<0.001). BMI showed a negative correlation with SI directional errors (R=-0.45, P<0.001), while age was not significantly correlated with the setup errors (P>0.05). CONCLUSIONS:The customized prone board demonstrates clinically significant advantages for its high setup accuracies in RL and SI directions in spite of its systematic AP directional errors. The setup accuracy in the SI direction is especially important for patients with lymph node irradiation or low BMI. Our findings provide quantitative evidence for immobilization device selection and individualized optimization of precision radiotherapy for prostate cancer.

OBJECTIVES:To propose a new method for optimizing radiotherapy planning for lung cancer by incorporating prognostic models that take into account individual patient information and assess the feasibility of treatment planning optimization directly guided by minimizing the predicted prognostic risk. METHODS:A mixed fluence map optimization objective was constructed, incorporating the outcome-based objective and the physical dose constraints. The outcome-based objective function was constructed as an equally weighted summation of prognostic prediction models for local control failure, radiation-induced cardiac toxicity, and radiation pneumonitis considering clinical risk factors. These models were derived using Cox regression analysis or Logistic regression. The primary goal was to minimize the outcome-based objective with the physical dose constraints recommended by the clinical guidelines. The efficacy of the proposed method for optimizing treatment plans was tested in 15 cases of non-small cell lung cancer in comparison with the conventional dose-based optimization method (clinical plan), and the dosimetric indicators and predicted prognostic outcomes were compared between different plans. RESULTS:In terms of the dosemetric indicators, D95% of the planning target volume obtained using the proposed method was basically consistent with that of the clinical plan (100.33% vs 102.57%, P=0.056), and the average dose of the heart and lungs was significantly decreased from 9.83 Gy and 9.50 Gy to 7.02 Gy (t=4.537, P<0.05) and 8.40 Gy (t=4.104, P<0.05), respectively. The predicted probability of local control failure was similar between the proposed plan and the clinical plan (60.05% vs 59.66%), while the probability of radiation-induced cardiac toxicity was reduced by 1.41% in the proposed plan. CONCLUSIONS:The proposed optimization method based on a mixed objective function of outcome prediction and physical dose provides effective protection against normal tissue exposure to improve the outcomes of lung cancer patients following radiotherapy.

BACKGROUND:The flat-panel x-ray source (FPXS) is an innovative x-ray source comprising millions of micro sources that emit low-energy x-rays. To date, the dosimetric characteristics of FPXS remain unclear, limiting its clinical applications. PURPOSE:This study aims to characterize the physical and dosimetric characteristics of FPXS through Monte Carlo (MC) dose calculation and experimental dosimetry measurements and to analyze the feasibility of FPXS as an electronic brachytherapy (EB) unit. METHODS:A measurement setup and two specialized measurement platforms were developed. By using an energy spectrometer, ionization chamber, and Gafchromic film, a measurement procedure for physical beam characteristics and dosimetric characteristics of FPXS was established, and a series of experimental measurements were conducted. A previously in-house developed MC simulation toolkit, designated gFPDMC, was employed to perform FPXS structure simulation and parallel dose calculation. The measured dose distributions were compared with the gFPDMC-calculated dose distributions, thereby validating the accuracy of gFPDMC. RESULTS:The measured energy spectra are consistent with the MC-simulated spectra with mean relative error (MRE) values < 9% and an average energy difference < 1.40%. The measured surface dose rate of FPXS increases exponentially with the operating voltage, with a maximum dose rate of 2.45 Gy/min at 40 kV; the depth dose rate decreases exponentially with the measurement depth and increases with the operating voltage. The doses calculated by gFPDMC, DOSXYZnrc, and Geant4 exhibit excellent consistency, with mean absolute error (MAE) values of 0.31% and 0.40%. The gFPDMC-calculated percentage depth dose (PDDs) exhibit identical voltage-dependent depth dose characteristics as measured PDDs, with deviations < 4%. CONCLUSIONS:The precision of gFPDMC was rigorously validated through comparison with DOSXYZnrc, Geant4, and experimentally measured dose. Adequate operating voltage and suitable dose rate demonstrate the feasibility of FPXS as an EB unit.

OBJECTIVES:To explore the synthesis of high-quality CT (sCT) from cone-beam CT (CBCT) using PE-CycleGAN for adaptive radiotherapy (ART) for nasopharyngeal carcinoma. METHODS:A perception-enhanced CycleGAN model "PE-CycleGAN" was proposed, introducing dual-contrast discriminator loss, multi-perceptual generator loss, and improved U-Net structure. CBCT and CT data from 80 nasopharyngeal carcinoma patients were used as the training set, with 7 cases as the test set. By quantifying the mean absolute error (MAE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), as well as the dose gamma pass rate and the relative dose deviations of the target area and organs at risk (OAR) between sCT and reference CT, the image quality and dose calculation accuracy of sCT were evaluated. RESULTS:The MAE of sCT generated by PE-CycleGAN compared to the reference CT was (56.89±13.84) HU, approximately 30% lower than CBCT's (81.06±15.86) HU (P<0.001). PE-CycleGAN's PSNR and SSIM were 26.69±2.41dB and 0.92±0.02 respectively, significantly higher than CBCT's 21.54±2.37dB and 0.86±0.05 (P<0.001), indicating substantial improvements in image quality and structural similarity. In gamma analysis, under the 2 mm/2% criterion, PE-CycleGAN's sCT achieved a pass rate of (90.13±3.75)%, significantly higher than CBCT's (81.65±3.92)% (P<0.001) and CycleGAN's (87.69±3.50)% (P<0.05). Under the 3 mm/3% criterion, PE-CycleGAN's sCT pass rate of (90.13±3.75)% was also significantly superior to CBCT's (86.92±3.51)% (P<0.001) and CycleGAN's (94.58±2.23)% (P<0.01). The mean relative dose deviation of the target area and OAR between sCT and planned CT was within ±3% for all regions, except for the Lens Dmax (Gy), which had a deviation of 3.38% (P=0.09). The mean relative dose deviations for PTVnx HI, PTVnd HI, PTVnd CI, PTV1 HI, PRV_SC, PRV_BS, Parotid, Larynx, Oral, Mandible, and PRV_ON were all less than ±1% (P>0.05). CONCLUSIONS:PE-CycleGAN demonstrates the ability to rapidly synthesize high-quality sCT from CBCT, offering a promising approach for CBCT-guided adaptive radiotherapy in nasopharyngeal carcinoma.

Artificial intelligence (AI) has been increasingly applied in cancer prevention, diagnosis, prognosis, treatment planning, and therapy implications. For enhancing professional communication and promoting research collaboration, Visualized Cancer Medicine continues the program of publishing the 150 most important questions in cancer research and clinical oncology. In this article, we propose three new key questions about integrating AI into radiation therapy for cancer patients as follows. Question 105: How can we develop individualized radiation therapy based on the biological variations combined with AI analysis for better treatment outcomes and less treatment toxicity? Question 106: Can AI improve real-time dose monitoring and adjustments in radiotherapy? Question 107: Can molecular profiling plus AI be help predict the benefits of adjusting the plan in adaptive radiotherapy?

Background At present, the implementation of intensity-modulated radiation therapy (IMRT) treatment planning for geometrically complex nasopharyngeal carcinoma (NPC) through manual trial-and-error fashion presents challenges to the improvement of planning efficiency and the obtaining of high-consistency plan quality. This paper aims to propose an automatic IMRT plan generation method through fluence prediction and further plan fine-tuning for patients with NPC and evaluates the planning efficiency and plan quality. Methods A total of 38 patients with NPC treated with nine-beam IMRT were enrolled in this study and automatically re-planned with the proposed method. A trained deep learning model was employed to generate static field fluence maps for each patient with 3D computed tomography images and structure contours as input. Automatic IMRT treatment planning was achieved by using its generated dose with slight tightening for further plan fine-tuning. Lastly, the plan quality was compared between automatic plans and clinical plans. Results The average time for automatic plan generation was less than 4 min, including fluence maps prediction with a python script and automated plan tuning with a C# script. Compared with clinical plans, automatic plans showed better conformity and homogeneity for planning target volumes (PTVs) except for the conformity of PTV-1. Meanwhile, the dosimetric metrics for most organs at risk (OARs) were ameliorated in the automatic plan, especially D max of the brainstem and spinal cord, and D mean of the left and right parotid glands significantly decreased ( P < 0.05). Conclusion We have successfully implemented an automatic IMRT plan generation method for patients with NPC. This method shows high planning efficiency and comparable or superior plan quality than clinical plans. The qualitative results before and after the plan fine-tuning indicates that further optimization using dose objectives generated by predicted fluence maps is crucial to obtain high-quality automatic plans.

Background Current intensity-modulated radiation therapy (IMRT) treatment planning is still a manual and time/resource consuming task, knowledge-based planning methods with appropriate predictions have been shown to enhance the plan quality consistency and improve planning efficiency. This study aims to develop a novel prediction framework to simultaneously predict dose distribution and fluence for nasopharyngeal carcinoma treated with IMRT, the predicted dose information and fluence can be used as the dose objectives and initial solution for an automatic IMRT plan optimization scheme, respectively. Methods We proposed a shared encoder network to simultaneously generate dose distribution and fluence maps. The same inputs (three-dimensional contours and CT images) were used for both dose distribution and fluence prediction. The model was trained with datasets of 340 nasopharyngeal carcinoma patients (260 cases for training, 40 cases for validation, 40 cases for testing) treated with nine-beam IMRT. The predicted fluence was then imported back to treatment planning system to generate the final deliverable plan. Predicted fluence accuracy was quantitatively evaluated within projected planning target volumes in beams-eye-view with 5 mm margin. The comparison between predicted doses, predicted fluence generated doses and ground truth doses were also conducted inside patient body. Results The proposed network successfully predicted similar dose distribution and fluence maps compared with ground truth. The quantitative evaluation showed that the pixel-based mean absolute error between predicted fluence and ground truth fluence was 0.53% ± 0.13%. The structural similarity index also showed high fluence similarity with values of 0.96 ± 0.02. Meanwhile, the difference in the clinical dose indices for most structures between predicted dose, predicted fluence generated dose and ground truth dose were less than 1 Gy. As a comparison, the predicted dose achieved better target dose coverage and dose hot spot than predicted fluence generated dose compared with ground truth dose. Conclusion We proposed an approach to predict 3D dose distribution and fluence maps simultaneously for nasopharyngeal carcinoma patients. Hence, the proposed method can be potentially integrated in a fast automatic plan generation scheme by using predicted dose as dose objectives and predicted fluence as a warm start.

Objective:To investigate the role of three-dimensional dose distribution-based deep learning model in predicting distant metastasis of head and neck cancer.Methods:Radiotherapy and clinical follow-up data of 237 patients with head and neck cancer undergoing intensity-modulated radiotherapy (IMRT) from 4 different institutions were collected. Among them, 131 patients from HGJ and CHUS institutions were used as the training set, 65 patients from CHUM institution as the validation set, and 41 patients from HMR institution as the test set. Three-dimensional dose distribution and GTV contours of 131 patients in the training set were input into the DM-DOSE model for training and then validated with validation set data. Finally, the independent test set data were used for evaluation. The evaluation content included the area under receiver operating characteristic curve (AUC), balanced accuracy, sensitivity, specificity, concordance index and Kaplan-Meier survival curve analysis.Results:In terms of prognostic prediction of distant metastasis of head and neck cancer, the DM-DOSE model based on three-dimensional dose distribution and GTV contours achieved the optimal prognostic prediction performance, with an AUC of 0.924, and could significantly distinguish patients with high and low risk of distant metastasis (log-rank test, P<0.001). Conclusion:Three-dimensional dose distribution has good predictive value for distant metastasis in head and neck cancer patients treated with IMRT, and the constructed prediction model can effectively predict distant metastasis.

Synthesizing computed tomography (CT) images from magnetic resonance imaging (MRI) data can provide the necessary electron density information for accurate dose calculation in the treatment planning of MRI-guided radiation therapy (MRIgRT). Inputting multimodality MRI data can provide sufficient information for accurate CT synthesis: however, obtaining the necessary number of MRI modalities is clinically expensive and time-consuming. In this study, we propose a multimodality MRI synchronous construction based deep learning framework from a single T1-weight (T1) image for MRIgRT synthetic CT (sCT) image generation. The network is mainly based on a generative adversarial network with sequential subtasks of intermediately generating synthetic MRIs and jointly generating the sCT image from the single T1 MRI. It contains a multitask generator and a multibranch discriminator, where the generator consists of a shared encoder and a splitted multibranch decoder. Specific attention modules are designed within the generator for feasible high-dimensional feature representation and fusion. Fifty patients with nasopharyngeal carcinoma who had undergone radiotherapy and had CT and sufficient MRI modalities scanned (5550 image slices for each modality) were used in the experiment. Results showed that our proposed network outperforms state-of-the-art sCT generation methods well with the least MAE, NRMSE, and comparable PSNR and SSIM index measure. Our proposed network exhibits comparable or even superior performance than the multimodality MRI-based generation method although it only takes a single T1 MRI image as input, thereby providing a more effective and economic solution for the laborious and high-cost generation of sCT images in clinical applications.

PurposeTo propose a novel magnetic field dose calculation method based on transformation from pencil beam (PB) to Monte Carlo (MC) distribution for MRI-Linac online treatment planning. MethodsThe novel magnetic field dose calculation algorithm was established by a PB dose engine and a magnetic field with tissue inhomogeneity influence correction network. The correction network was constructed with a Res-UNet framework, including residual modules and an encoding-decoding path, by inputting three-dimensional PB dose and patient electron density map, and outputting transformed dose distribution. The influences of magnetic fields and tissue heterogeneity were considered and corrected simultaneously in the correction model. A total of 110 clinically treated static beam IMRT plans were collected, including plans for brain, head-and-neck, lung, and rectum cases. A total of 90 cases were used and enhanced to train and validate the model, and the other 20 cases were for test. By comparing the proposed pipeline-generated dose distribution with original input PB dose and corresponding MC dose, the feasibility and effectiveness of the method was evaluated. ResultsResults on both beam dose and plan dose accuracy comparisons on all investigated four tumor sites show great consistency between the cross-dose-engine transformation generations and the MC results, with averaged plan mean absolute error of 0.90% +/- 0.13% for the voxel-wise dose difference and 98.33% +/- 1.07% gamma passing rate at the 2%/2 mm criteria. The whole PB calculation and transformation process can be completed within second. ConclusionsWe have successfully developed a fast novel magnetic field dose calculation pipeline based on transformation from PB distribution to MC distribution for MRI-Linac online treatment planning.

In knowledge-based treatment planning (KBTP) for intensity-modulated radiation therapy (IMRT), the quality of the plan is dependent on the sophistication of the predicted dosimetric information and its application. In this paper, we propose a KBTP method that based on the effective and reasonable utilization of a three-dimensional (3D) dose prediction on planning optimization. We used an organs-at-risk (OARs) dose distribution prediction model to create a voxel-based dose sequence based optimization objective for OARs doses. This objective was used to reformulate a traditional fluence map optimization model, which involves a tolerable spatial re-assignment of the predicted dose distribution to the OAR voxels based on their current doses' positions at a sorted dose sequencing. The feasibility of this method was evaluated with ten gynecology (GYN) cancer IMRT cases by comparing its generated plan quality with the original clinical plan. Results showed feasible plan by proposed method, with comparable planning target volume (PTV) dose coverage and greater dose sparing of the OARs. Among ten GYN cases, the average V30 and V45 of rectum were decreased by 4%±4% (p = 0.02) and 4%±3% (p<0.01), respectively. V30 and V45 of bladder were decreased by 8%±2% (p<0.01) and 3%±2% (p<0.01), respectively. Our predicted dose sequence-based planning optimization method for GYN IMRT offered a flexible use of predicted 3D doses while ensuring the output plan consistency.