Radiotherapy (RT) represents a core treatment modality in oncology. Its workflow encompasses multiple critical stages: patient immobilization and positioning, imaging simulation, target contouring, treatment planning, quality assurance (QA), and treatment delivery in sequence. Despite technological advances, many steps remain heavily dependent on manual operations, leading to human-induced uncertainties and limited efficiency and standardization. In this context, artificial intelligence (AI) has emerged as a transformative force in radiation oncology, providing innovative solutions to automate repetitive tasks, enhance clinical decision-making, and personalize treatment strategies. This study presents a comprehensive review of AI applications across three core domains of the RT workflow: (1) automatic segmentation of organs at risk (OARs) and tumor targets, (2) AI-driven treatment planning, including dose prediction and optimization, and (3) AI-assisted quality assurance, encompassing patient-specific, machine-specific, and outcome-based evaluations. For each domain, it presents a summary of representative algorithmic strategies, evaluates their clinical performance, and discusses current limitations. Beyond task-specific automation, this study examines the clinical integration of AI tools throughout the entire RT workflow, including imaging-oriented AI, conversational AI, and large language model (LLM) applications. Furthermore, this study critically analyzes the technical, clinical, and regulatory challenges that impede widespread AI adoption, including limited model generalizability, poor interpretability, data heterogeneity, automation bias, and a lack of standardized validation frameworks. In addition, this study explores the evolving roles of human-AI collaboration, safety considerations, and responsible deployment strategies in clinical practice. Finally, this study outlines future directions for AI in RT, emphasizing the need for multicenter validation, workflow-level evaluations, and interdisciplinary collaboration to translate AI's technical promise into tangible improvements in patient care. The ultimate goal of AI application in RT is to augment the cognitive capabilities of clinicians rather than replace them, thus enabling more precise, efficient, and equitable cancer treatment worldwide.

This study compared the plan quality and execution efficiency of VMAT plans using TrueBeam single-isocenter and Halcyon dual-isocenter systems for postoperative left-sided breast cancer, providing evidence for the selection and application of radiotherapy methods after breast cancer surgery. Twenty postoperative left-sided breast cancer patients treated at Zhejiang Cancer Hospital from January 2020 to August 2022 were included in the study. Using the RayStation 9 A system, TrueBeam single-isocenter (T group) and Halcyon dual-isocenter (H group) VMAT plans were created. The study compared PTV (Planning Target Volume) conformity index (CI), homogeneity index (HI), mean and maximum doses, dose to organs at risk (OAR), monitor units (MU), beam-on time, and plan complexity between the groups. Both groups’ plans met clinical objectives with good dose distribution and target coverage. Significant statistical differences were observed in the PTV maximum dose (Dmax), mean dose (Dmean), and D50 between the two groups (P < 0.05), with the H group outperforming the T group. There was no statistically significant difference in CI and HI between the two groups (P > 0.05). The OAR doses were lower in the H group compared to the T group, with significant statistical differences observed in heart Dmean, left lung (Dmean, V5, V20, V30), lungs (Dmean, V5, V20, V30), and humeral head Dmax (P < 0.05), and there was no statistically significant difference in heart (V30, V40), spinal cord Dmax, and right breast Dmean (P < 0.05). In terms of plan quality assessment, the total MU for the H group was significantly higher (1066.704 ± 135.183) than that for the T group (772.992 ± 104.023), with extremely significant statistical differences (P < 0.01). The beam-on time for the H group was significantly longer (101.328 ± 9.359 s) than that for the T group (53.535 ± 12.655 s), with extremely significant statistical differences (P < 0.01). The plan complexity for the H group was significantly higher (0.101 ± 0.009 cm−1) than that for the T group (0.068 ± 0.009 cm−1), with extremely significant statistical differences (P < 0.01). The study found that Halcyon dual-isocenter VMAT plans offered better PTV uniformity and coverage while protecting OARs more effectively. Despite higher MU, beam-on time, and plan complexity, the H group’s dosimetric advantages and reduced manual positioning errors make it a safe and effective treatment option for postoperative breast cancer.

BACKGROUND:Four-dimension computed tomography (4D-CT) provides important respiration-related information for thoracic radiotherapy. Its quality is challenged by various respiratory patterns. Its acquisition gives rise to the risk of higher radiation exposure. Based on a continuously estimated deformation, a 4D synthesis by warping a high-quality volumetric image is a possible solution. PURPOSE:To propose a non-patient-specific cascaded ensemble model (CEM) to estimate respiration-induced thoracic tissue deformation from surface motion. METHODS:The CEM is cascaded by three deep learning-based models. By inputting the surface motion, CEM outputs a deformation vector field (DVF) inside thorax. In our work, the surface motion was simulated using the body contours derived from 4D-CT. The CEM was trained on our private database including 62 4D-CT sets, and was tested on a public database encompassing 80 4D-CT sets. To evaluate CEM, we employed the model output DVF to generate a few series of synthesized CTs, and compared them with the ground truth. CEM was also compared with other published works. RESULTS:CEM synthesized CT with an mRMSE (average root mean square error) of 61.06 ± 10.43HU (average ± standard deviation), an mSSIM (average structural similarity index measure) of 0.990 ± 0.004, and an mMAE (average mean absolute error) of 26.80 ± 5.65HU. Compared with other works, CEM showed the best result. CONCLUSIONS:The results demonstrated the effectiveness of CEM on estimating tissue DVF inside thorax. CEM requires no patient-specific breathing data sampling and no additional training before treatment. It shows potential for broad applications.

Deep learning (DL) -based automated treatment planning (ATP) shows significant promise in streamlining radiotherapy workflow and reducing variability in plan quality. However, it often lacks the flexibility needed for achieving individualized trade-offs in real-world practice. Herein, we propose a hybrid strategy by integrating DL-based dose prediction with clinical-goal-guided inverse optimization to generate directly deliverable plans within five minutes. DL models for five disease sites were trained separately using datasets from a single institution and were tested retrospectively for clinical application among three institutions, with tailored prioritized clinical goals. We find that over 80% of the 250 auto-plans met clinical criteria, and 60% were preferred over manual plans in blinded reviews. Dosimetric analyses show that the auto-plans quantitatively matched or exceeded the quality of human-driven plans. This study highlights ATP's potential to transform radiotherapy practice, with ongoing efforts aimed at refining its versatility and adoption across diverse clinical settings.

Accurate and efficient automatic segmentation is essential for various clinical tasks such as radiotherapy treatment planning. However, atlas-based segmentation still faces challenges due to the lack of representative atlas dataset and the computational limitations of deformation algorithms. In this work, we have proposed an atlas selection procedure (subset atlas grouping approach, MAS-SAGA) which utilized both image similarity and volume features for selecting the best-fitting atlases for contour propagation. A dataset of anonymized female pelvic Computed Tomography (CT) images demonstrated that MAS-SAGA significantly outperforms conventional multi-atlas-based segmentation (cMAS) in terms of Dice Similarity Coefficient (DSC) and 95th Percentile Hausdorff Distance (95HD) for bladder and rectum segmentation using a three-fold cross-validation strategy. The proposed procedure also reduced computation time compared to cMAS, making it a promising tool for medical image analysis applications. In addition, we have evaluated two distinct atlas selection methods: the Feature-based Atlas Selection Approach (MAS-FASA) and the Similarity-based Atlas Selection Approach (MAS-SIM). We investigate the differences between these two methods in terms of their ability to select the best fitting atlases. The findings demonstrated that MAS-FASA selected different atlases than MAS-SIM, resulting in improved segmentation performance overall. It highlighted the potential of feature-based subgrouping techniques in enhancing the efficacy of MAS algorithms in the field of medical image segmentation.

This study aims to explore the prognostic value of regionally modulated radiomics for patients with head and neck cancer (HNC) in positron emission tomography/computed tomography (PET/CT) imaging. The dataset included 224 HNC patients who underwent PET/CT imaging at five different centers. The primary tumor was manually contoured by experienced radiologists. For introducing regionally modulated radiomics, we developed four fuzzy masks by applying Gaussian filter, and four peritumor-included masks by applying morphological operations. For each patient, a total of 326 radiomic features were extracted from each of nine masks. Multivariate Cox proportional hazards model with ensemble strategy was adopted to construct classical, fuzzy, and peritumoral based prognostic models, respectively, for predicting progression-free survival. ComBat harmonization was applied to adjust for multicenter variability. A consistent modelling approach was employed to ensure the independence and comparability of these models. The models were evaluated by C-index, log-rank test, and the area under the time-dependent ROC curve (tAUC). The fuzzy radiomics model applied with 5 mm FWHM of Gaussian filter demonstrated superior performance compared to classical radiomics model (Testing C-index, 0.735 vs. 0.685; log-rank test, p < 0.007 vs. p < 0.035). Peritumoral radiomics models showed slightly improved performance compared to classical radiomics model (Testing C-index, 0.727 vs. 0.685; log-rank test, p < 0.014 vs. p < 0.035). The tAUC demonstrated consistent findings with the C-index. The harmonization strategy showed further improved performance for both fuzzy and peritumoral models. These results showed that regionally modulated radiomics analysis was superior for estimating prognosis in this multicenter HNC cohort when compared to classical radiomics. This demonstrated the potentially prognostic values by considering regional variations in radiomics analysis.

The purpose of this study is to develop an electronic portal imaging device‐based multi‐leaf collimator calibration procedure using log files. Picket fence fields with 2–14 mm nominal strip widths were performed and normalized by open field. Normalized pixel intensity profiles along the direction of leaf motion for each leaf pair were taken. Three independent algorithms and an integration method derived from them were developed according to the valley value, valley area, full‐width half‐maximum (FWHM) of the profile, and the abutment width of the leaf pairs obtained from the log files. Three data processing schemes (Scheme A, Scheme B, and Scheme C) were performed based on different data processing methods. To test the usefulness and robustness of the algorithm, the known leaf position errors along the direction of perpendicular leaf motion via the treatment planning system were introduced in the picket fence field with nominal 5, 8, and 11 mm. Algorithm tests were performed every 2 weeks over 4 months. According to the log files, about 17.628% and 1.060% of the leaves had position errors beyond ± 0.1 and ± 0.2 mm, respectively. The absolute position errors of the algorithm tests for different data schemes were 0.062 ± 0.067 (Scheme A), 0.041 ± 0.045 (Scheme B), and 0.037 ± 0.043 (Scheme C). The absolute position errors of the algorithms developed by Scheme C were 0.054 ± 0.063 (valley depth method), 0.040 ± 0.038 (valley area method), 0.031 ± 0.031 (FWHM method), and 0.021 ± 0.024 (integrated method). For the efficiency and robustness test of the algorithm, the absolute position errors of the integration method of Scheme C were 0.020 ± 0.024 (5 mm), 0.024 ± 0.026 (8 mm), and 0.018 ± 0.024 (11 mm). Different data processing schemes could affect the accuracy of the developed algorithms. The integration method could integrate the benefits of each algorithm, which improved the level of robustness and accuracy of the algorithm. The integration method can perform multi‐leaf collimator (MLC) quality assurance with an accuracy of 0.1 mm. This method is simple, effective, robust, quantitative, and can detect a wide range of MLC leaf position errors.

Background and Purpose Artificial intelligence (AI) is a technique which tries to think like humans and mimic human behaviors. It has been considered as an alternative in a lot of human-dependent steps in radiotherapy (RT), since the human participation is a principal uncertainty source in RT. The aim of this work is to provide a systematic summary of the current literature on AI application for RT, and to clarify its role for RT practice in terms of clinical views. Materials and Methods A systematic literature search of PubMed and Google Scholar was performed to identify original articles involving the AI applications in RT from the inception to 2022. Studies were included if they reported original data and explored the clinical applications of AI in RT. Results The selected studies were categorized into three aspects of RT: organ and lesion segmentation, treatment planning and quality assurance. For each aspect, this review discussed how these AI tools could be involved in the RT protocol. Conclusions Our study revealed that AI was a potential alternative for the human-dependent steps in the complex process of RT.

Introduction Radiotherapy (RT) is the main treatment for patients with nasopharyngeal carcinoma (NPC). NPC patients at different stages have varying levels of damage to normal brain tissue after RT. No study has yet thoroughly analyzed the variations in radiation dosages in the brain for different stages of NPC patients treated with RT. This study aims to examine these variations. Methods 1446 NPC patients’ CT and RTdose data were retrospectively reviewed. Analysis of the radiation dosage was executed on these 803 patients. The RTdose images for several patient groups were averaged after registering each patient’s RTdose data to the CT brain template created in our earlier study. The voxel-based (VB) analysis was used to examine the dose variations in the brains of three groups of NPC patients: the early-stage group, the stage III group, and the stage IV group. Results As the disease progresses from early to advanced stages, the intensity and volume of radiation in the brain increase. The normal brain tissue accepted a substantially larger dosage in more advanced NPC patients. Differences in brain regions between stage III and early-stage patients were minimal compared to any other two groups. Brain regions exhibited substantial variations between the stage IV group and all other patient groups were broadly distributed. Conclusion Our findings highlight the critical role of NPC staging in the therapeutic strategy, emphasizing the heterogeneity of radiation-induced tissue damage across disease stages and implying the need to develop stage-specific RT plans.

Background: To determine whether a dual-isocenter volumetrically modulated arc therapy (VMAT) technique results in lower normal pulmonary dosage compared to a traditional single isocenter technique for boot-shaped lung cancer. Methods: A cohort of 15 patients with advanced peripheral or central lung cancer who had metastases in the mediastinum and supraclavicular lymph nodes was randomly selected for this retrospective study. VMAT plans were generated for each patient using two different beam alignment techniques with the 6-MV flattening filter-free (FFF) photon beam: single-isocenter jaw-tracking VMAT based on the Varian TrueBeam linear accelerator (S-TV), and dual-isocenter VMAT based on both TrueBeam (D-TV) and Halcyon linear accelerator (D-HV). For all 45 treatment plans, planning target volume (PTV) dose coverage, conformity/homogeneity index (CI/HI), mean heart dose (MHD), mean lung dose (MLD) and the total lung tissue receiving 5, 20, 30 Gy (V-5, V-20, V-30) were evaluated. The monitor units (MUs), delivery time, and plan quality assurance (QA) results were recorded. Results: The quality of the objectives of the three plans was comparable to each other. In comparison with S-TV, D-TV and D-HV improved the CI and HI of the PTV (p < 0.05). The MLD was 13.84 +/- 1.44 Gy (mean +/- SD) for D-TV, 14.22 +/- 1.30 Gy and 14.16 +/- 1.42 Gy for S-TV and D-HV, respectively. Lungs-V-5Gy was 50.78 +/- 6.24%, 52.00 +/- 7.32% and 53.36 +/- 8.48%, Lungs-V-20Gy was 23.72 +/- 2.27%, 26.18 +/- 2.86% and 24.96 +/- 3.09%, Lungs-V30Gy was 15.69 +/- 1.76%, 17.20 +/- 1.72% and 16.52 +/- 2.07%. Compared to S-TV, D-TV provided statistically significant better protection for the total lung, with the exception of the lungs-V-5. All plans passed QA according the gamma criteria of 3%/3 mm. Conclusions: Taking into account the dosimetric results and published clinical data on radiation-induced pulmonary injury, dual-isocenter jaw-tracking VMAT may be the optimal choice for treating boot-shaped lung cancer.

In imaged-guided radiation therapy (IGRT), target localization is usually done with rigid-body registration based on anatomy matching. Problems arise when the target volume can only be matched partially due to interfractional organ motion and deformation, resulting in deteriorated target coverage and critical structure sparing. A new target localization method is investigated in which the treatment target volume is aligned with the prescription isodose surface. Our study included 15 prostate patients previously treated with intensitymodulated radiation therapy (IMRT). Patient setup and target localization were performed using a CT-on-rails system before and after the IMRT treatment. IMRT plans were generated on the original simulation CTs (15) and the same MUs and leaf sequences were used to compute the dose distributions on post-treatment CTs (98) with the isocenter adjustments based on either anatomical structure matching or prescription isodose surface alignment. When patients were aligned with the traditional anatomy matching method, the dose to 95% of the CTV, D95, received 74.0 - 77.6 Gy and the minimum CTV dose, Dmin, was 61.9 - 71.6 Gy, respectively, in the cumulative dose distributions. The rectal dose-volume constraints were violated in 35.7% of the treatment fractions. When patients were aligned using the new localization method, the dose to 95% of the CTV, D95, received 74.0 - 78.2 Gy and the minimum CTV dose, Dmin, was 68.4 - 71.6 Gy, respectively, in the cumulative dose distributions. The rectal dose-volume constraints were violated in 17.3% of the treatment fractions. Traditional IGRT target localization based on anatomy matching is effective for population-based PTV margins but not ideal for those patients with large inter-fractional prostate rotation/deformation due to large rectal and bladder volume variation. The new method using the prescription isodose surface to align the target volume could improve the target coverage and rectal sparing for these patients, which can be implemented clinically to improve target dose delivery accuracy.

Background: Cone-beam computed tomography (CBCT) is an important tool for patient positioning in radiotherapy due to its outstanding advantages. However, the CBCT registration shows errors due to the limitations of the automatic registration algorithm and the nonuniqueness of manual verification results. The purpose of this study was to verify the feasibility of using the Sphere-Mask Optical Positioning System (S-M_ OPS) to improve the registration stability of CBCT through clinical trials. Methods: From November 2021 to February 2022, 28 patients who received intensity-modulated radiotherapy and site verification with CBCT were included in this study. S- M_OPS was used as an independent third-party system to supervise the CBCT registration result in real time. The supervision error was calculated based on the CBCT registration result and using the S-M_OPS registration result as the standard. For the head and neck, patients with a supervision error >= 3 or <=-3 mm in 1 direction were selected. For the thorax, abdomen, pelvis, or other body parts, patients with a supervision error >= 5 or <=-5 mm in 1 direction were selected. Then, re-registration was performed for all patients (selected and unselected). The registration errors of CBCT and S-M_OPS were calculated based on the re-registration results as the standard. Results: For selected patients with large supervision errors, CBCT registration errors (mean +/- standard deviation) in the latitudinal (LAT; left/right), vertical (VRT; superior/inferior), and longitudinal (LNG; anterior/posterior) directions were 0.90 +/- 3.20, -1.70 +/- 0.98, and 7.30 +/- 2.14 mm, respectively. The S-M_OPS registration errors were 0.40 +/- 0.14, 0.32 +/- 0.66, and 0.24 +/- 1.12 mm in the LAT, VRT, and LNG directions, respectively. For all patients, CBCT registration errors in the LAT, VRT, and LNG directions were 0.39 +/- 2.69, -0.82 +/- 1.47, and 2.39 +/- 2.93 mm, respectively. The S-M_OPS registration errors were -0.25 +/- 1.33, 0.55 +/- 1.27, and 0.36 +/- 1.34 mm for all patients in the LAT, VRT, and LNG directions, respectively. Conclusions: This study shows that S-M_OPS registration offers comparable accuracy to CBCT for daily registration. S-M_OPS, as an independent third-party tool, can prevent large errors in CBCT registration, thereby improving the accuracy and stability of CBCT registration.

Background The machine learning models with dose factors and the deep learning models with dose distribution matrix have been used to building lung toxics models for radiotherapy and achieve promising results. However, few studies have integrated clinical features into deep learning models. This study aimed to explore the role of three-dimension dose distribution and clinical features in predicting radiation pneumonitis (RP) in esophageal cancer patients after radiotherapy and designed a new hybrid deep learning network to predict the incidence of RP. Methods A total of 105 esophageal cancer patients previously treated with radiotherapy were enrolled in this study. The three-dimension (3D) dose distributions within the lung were extracted from the treatment planning system, converted into 3D matrixes and used as inputs to predict RP with ResNet. In total, 15 clinical factors were normalized and converted into one-dimension (1D) matrixes. A new prediction model (HybridNet) was then built based on a hybrid deep learning network, which combined 3D ResNet18 and 1D convolution layers. Machine learning-based prediction models, which use the traditional dosiomic factors with and without the clinical factors as inputs, were also constructed and their predictive performance compared with that of HybridNet using tenfold cross validation. Accuracy and area under the receiver operator characteristic curve (AUC) were used to evaluate the model effect. DeLong test was used to compare the prediction results of the models. Results The deep learning-based model achieved superior prediction results compared with machine learning-based models. ResNet performed best in the group that only considered dose factors (accuracy, 0.78 ± 0.05; AUC, 0.82 ± 0.25), whereas HybridNet performed best in the group that considered both dose factors and clinical factors (accuracy, 0.85 ± 0.13; AUC, 0.91 ± 0.09). HybridNet had higher accuracy than that of Resnet ( p = 0.009). Conclusion Based on prediction results, the proposed HybridNet model could predict RP in esophageal cancer patients after radiotherapy with significantly higher accuracy, suggesting its potential as a useful tool for clinical decision-making. This study demonstrated that the information in dose distribution is worth further exploration, and combining multiple types of features contributes to predict radiotherapy response.

OBJECTIVE:By comparing the target dose distribution with or without the robust optimization, the dosimetric advantages of robust optimization and flattening filter free (FFF) in radiation therapy for postmastectomy cancer of the left breast was explored when part of the chest wall target was moved out in case of respiratory motion. MATERIALS AND METHODS:This is a retrospective study. The data of 21 postmastectomy patients with cancer of the left breast from 2019 to 2020 were retrospectively collected. The planned target volume (PTV) dose was prescribed 50 Gy/25 fractions and the treatment plans were designed using 6 MV FFF X ray and volumetric modulated arc therapy (VMAT) technology in RayStation treatment planning system (TPS), with and without robust optimization. The movement of the target area of the internal chest wall (0.50 cm) caused by respiratory movement was simulated by moving the isocenter of the beams. RESULTS:When the chest wall target moved outward, the PTV target area D98, D95, D2, conformity index (CI) and homogeneity index (HI) with robust optimization were better than those without robust optimization. The coverage rate of Planned Target Volume-Chest (PTV-T) V50 with robust optimization was significantly higher than that with no-robust optimization (P<0.001). Clinical target volume (CTV) V50 coverage with robust optimization was 14.49% higher than that with no-robust optimization. In terms of organ-at-risk parameters, the average spinal cord dose of the plan with robust optimization was 13.19% lower than that of the plan with no-robust optimization, and the Lung-L V5 of the plan with no-robust optimization was slightly (1.94%) lower than that of the plan with robust optimization. There was no significant difference in machine execution efficiency between the two groups (P>0.05). CONCLUSIONS:Robust optimization could be adopted in the postoperative radiotherapy planning for cancer in the left breast, for it ensures that the target dose coverage and the dose limit of organ-at-risk still meet the clinical requirements under condition of chest wall displacement caused by respiratory movement.

Radiotherapy is a cancer treatment that uses high doses of radiation to kill cancer cells. Segmentation of the clinical target volumes (CTVs) and organs at risk (OARs) is an essential step in rectal cancer radiotherapy treatment planning. However, the manual segmentation of CTVs and OARs is labor-intensive and prone to intraand inter-rater variations. The challenge of this task lies in large shape variations and low-contrast structure boundary for some structures like intestine. The contextual information is very important to handle this challenge. Recently, deep learning has greatly improved the state-of-the-art results in automatic segmentation of anatomical structures including OARs and CTVs. However, existing approaches barely take advantages of relations among anatomical structures as contextual information, which limits the segmentation accuracy. In this study, we propose a novel structure-contextual representations approach based on 3D high-resolution network (3D HRNet-SCR) for segmentation of OARs and CTV of rectal cancer. First, we design a structure-contextual representation module (SCR) which could compute the representation for each structure of interest by aggregating the representations of voxels inside the structure and enhance the feature representation of each voxel with learned structure-contextual representations. Second, we extend the powerful 2D HRNet for semantic segmentation to 3D paradigm to better capture the contextual information across slices of volumetric data. Our propose approach integrates the proposed SCR module on top of the 3D HRNet to form a high-performance segmentation framework. Finally, we collect a large-scale rectal cancer dataset of 536 CT scans (total 64320 slices) for evaluation. Our proposed framework is extensively tested on this self-collected dataset, showing superiority compared with state-of-the-art OARs and CTV segmentation methods for rectal cancer treatment.

PURPOSE:A small animal radiation research platform (SARRP) equipped with a miniature beam system, an image-guided positioning system, and a dose planning system was used to develop and evaluate a mouse model of radiation-induced temporomandibular damage. METHODS:Left jaw disks of adult male C57BL/6 mice and C3H mice were targeted using the SARRP for image-guided irradiation. The total radiation dose was 75 Gy. Experiment 1 (Scoping study): Mice in the C57BL/6 mouse test and control groups were sacrificed at 1, 3, 6, 9, 12, 15, and 18 weeks after irradiation, whereas mice in the C3H test and control groups were sacrificed at 1, 3, 6, 9, and 12 weeks after irradiation. Experiment 2 (Full -scale validation study): Mice in the C57BL/6 mouse test and control groups were sacrificed at 1, 3 and 6 weeks after irradiation. Histopathological analysis of the temporomandibular skeletal muscle in each group was performed using hematoxylin and eosin (H&E) and Masson staining; the temporal mandibular bone was examined through H&E staining. RESULTS:SARRP delivered the rated dose to the temporomandibular joints of C57BL/6 and C3H mice. C3H and C57BL/6 mice in the test group showed different degrees of osteocytic necrosis and osteoporosis at different time points. H&E staining of skeletal muscle tissue showed slight fibrosis in the C57BL/6 test at 3 and 6 weeks time point. CONCLUSION:We established a model of radiation-induced damage in the temporomandibular joint of C57BL/6 mice and demonstrated that the observed physiological and histological changes correspond to radiation damage observed in humans. Furthermore, the SARRP can deliver precise radiation doses.

Background:The setup accuracy plays an extremely important role in the local control of tumors. The purpose of this study is to verify the feasibility of "Sphere-Mask" Optical Positioning System (S-M_OPS) for fast and accurate setup.Methods:From 2016 to 2021, we used S-M_OPS to supervise 15441 fractions in 1981patients (with the cancer in intracalvarium, nasopharynx, esophagus, lung, liver, abdomen or cervix) undergoing intensity-modulated radiation therapy (IMRT), and recorded the data such as registration time and mask deformation. Then, we used S-M_OPS, laser line and cone beam computed tomography (CBCT) for co-setup in 277 fractions, and recorded laser line-guided setup errors and S-M_OPS-guided setup errors with CBCT-guided setup result as the standard.Results:S-M_OPS supervision results: The average time for laser line-guided setup was 31.75s. 12.8% of the reference points had an average deviation of more than 2 mm and 5.2% of the reference points had an average deviation of more than 3 mm. Co-setup results: The average time for S-M_OPS-guided setup was 7.47s, and average time for CBCT-guided setup was 228.84s (including time for CBCT scan and manual verification). In the LAT (left/right), VRT (superior/inferior) and LNG (anterior/posterior) directions, laser line-guided setup errors (mean±SD) were -0.21±3.13mm, 1.02±2.76mm and 2.22±4.26mm respectively; the 95% confidence intervals (95% CIs) of laser line-guided setup errors were -6.35 to 5.93mm, -4.39 to 6.43mm and -6.14 to 10.58mm respectively; S-M_OPS-guided setup errors were 0.12±1.91mm, 1.02±1.81mm and -0.10±2.25mm respectively; the 95% CIs of S-M_OPS-guided setup errors were -3.86 to 3.62mm, -2.53 to 4.57mm and -4.51 to 4.31mm respectively.Conclusion:S-M_OPS can greatly improve setup accuracy and stability compared with laser line-guided setup. Furthermore, S-M_OPS can provide comparable setup accuracy to CBCT in less setup time.

Purpose: To propose a multi-output fully convolutional network (MOFCN) to segment bilateral lung, heart and spinal cord in the planning thoracic computed tomography (CT) slices automatically and simultaneously. Methods: The MOFCN includes two components: one main backbone and three branches. The main backbone extracts the features about lung, heart and spinal cord. The extracted features are transferred to three branches which correspond to three organs respectively. The longest branch to segment spinal cord is nine layers, including input and output layers. The MOFCN was evaluated on 19,277 CT slices from 966 patients with cancer in the thorax. In these slices, the organs at risk (OARs) were delineated and validated by experienced radiation oncologists, and served as ground truth for training and evaluation. The data from 61 randomly chosen patients were used for training and validation. The remaining 905 patients’ slices were used for testing. The metric used to evaluate the similarity between the auto-segmented organs and their ground truth was Dice. Besides, we compared the MOFCN with other published models. To assess the distinct output design and the impact of layer number and dilated convolution, we compared MOFCN with a multi-label learning model and its variants. By analyzing the not good performances, we suggested possible solutions. Results: MOFCN achieved Dice of 0.95 ± 0.02 for lung, 0.91 ± 0.03 for heart and 0.87 ± 0.06 for spinal cord. Compared to other models, MOFCN could achieve a comparable accuracy with the least time cost. Conclusion: The results demonstrated the MOFCN’s effectiveness. It uses less parameters to delineate three OARs simultaneously and automatically, and thus shows a relatively low requirement for hardware and has potential for broad application.

Fully convolutional networks were developed for predicting optimal dose distributions for patients with left-sided breast cancer and compared the prediction accuracy between two-dimensional and three-dimensional networks. Sixty cases treated with volumetric modulated arc radiotherapy were analyzed. Among them, 50 cases were randomly chosen to conform the training set, and the remaining 10 were to construct the test set. Two U-Net fully convolutional networks predicted the dose distributions, with two-dimensional and three-dimensional convolution kernels, respectively. Computed tomography images, delineated regions of interest, or their combination were considered as input data. The accuracy of predicted results was evaluated against the clinical dose. Most types of input data retrieved a similar dose to the ground truth for organs at risk ( p > 0.05). Overall, the two-dimensional model had higher performance than the three-dimensional model ( p < 0.05). Moreover, the two-dimensional region of interest input provided the best prediction results regarding the planning target volume mean percentage difference (2.40 ± 0.18%), heart mean percentage difference (4.28 ± 2.02%), and the gamma index at 80% of the prescription dose are with tolerances of 3 mm and 3% (0.85 ± 0.03), whereas the two-dimensional combined input provided the best prediction regarding ipsilateral lung mean percentage difference (4.16 ± 1.48%), lung mean percentage difference (2.41 ± 0.95%), spinal cord mean percentage difference (0.67 ± 0.40%), and 80% Dice similarity coefficient (0.94 ± 0.01). Statistically, the two-dimensional combined inputs achieved higher prediction accuracy regarding 80% Dice similarity coefficient than the two-dimensional region of interest input (0.94 ± 0.01 vs 0.92 ± 0.01, p < 0.05). The two-dimensional data model retrieves higher performance than its three-dimensional counterpart for dose prediction, especially when using region of interest and combined inputs.

BACKGROUND:Image-guided adaptive brachytherapy shows the ability to deliver high doses to tumors while sparing normal tissues. However, interfraction dose delivery introduces uncertainties to high dose estimation, which relates to normal tissue toxicity. The purpose of this study was to investigate the high-dose regions of two applicator approaches in brachytherapy.METHOD:For 32 cervical cancer patients, the CT images from each fraction were wrapped to a reference image, and the displacement vector field (DVF) was calculated with a hybrid intensity-based deformable registration algorithm. The fractional dose was then accumulated to calculate the position and the overlap of high dose (D2cc) during multiple fractions.RESULT:The overall Dice similarity coefficient (DSC) of the deformation algorithm for the bladder and the rectum was (0.97 and 0.91). No significant difference was observed between the two applicators. However, the location of the intracavitary brachytherapy (ICBT) high-dose region was relatively concentrated. The overlap volume of bladder and rectum D2cc was 0.42 and 0.71, respectively, which was higher than that of interstitial brachytherapy (ISBT) (0.26 and 0.31). The cumulative dose was overestimated in ISBT cases when using the GEC-recommended method. The ratio of bladder and rectum D2cc to the GEC method was 0.99 and 1, respectively, which was higher than that of the ISBT method (0.96 and 0.94).CONCLUSION:High-dose regions for brachytherapy based on different applicator types were different. The 3D-printed ICBT has better high-dose region consistency than freehand ISBT and hence is more predictable.