Objective:To assess intrafractional errors of the deformation of clinical target volumes (CTVs) during online adaptive radiotherapy (OART) for prostate cancer patients, aiming to provide a basis for online plan optimization.Methods:A retrospective analysis was conducted for 13 prostate cancer patients who received 1.5 T magnetic resonance imaging (MRI)-guided OART (8 Gy × 5 fractions, totaling 65 fractions) at the Department of Radiation Oncology, Cancer Hospital, Chinese Academy of Medical Sciences. The MRI images were collected at the beginning and end of various treatment fractions. Then, the CTVs and organs at risk (OARs) were delineated by the same radiation oncologist. After rigid registration and triangle mesh generation, the surface vertices were extracted. The deformable registration for the CTV surfaces was performed using the thin-plate spline robust point matching (TPS-RPM) algorithm, yielding vertex correspondences. Last, both systematic and random intrafractional errors of CTV deformation were calculated.Results:The average Hausdorff distance (HD) for deformable registration of treatment fractions was (1.68 ± 0.28) mm. The intrafractional systematic errors of CTV deformation were (0.25 ± 3.18) mm (anterior-posterior direction; A-P), (0.89 ± 3.85) mm (left-right direction, L-R), and (-1.98 ± 6.69) mm (superior-inferior direction, S-I). The intrafractional random errors of CTV deformation were determined at (-0.26 ± 1.89) mm (A-P), (-0.08 ± 0.88) mm (L-R), and (-0.04 ± 1.86) mm (S-I).Conclusions:During OART, CTV deformations primarily occur in the S-I direction. Therefore, it is necessary to consider the expanded size of margins in this direction during OART for prostate cancer.

Objective:To construct a general deep learning dose prediction model applicable to radiotherapy for head and neck tumors, establish design methods for artificial intelligence (AI)-assisted radiotherapy plan and evaluate the accuracy of prediction.Methods:Radiotherapy plans of 818 patients who received radiotherapy for head and neck cancers from January 2018 to June 2021 in Cancer Hospital of Chinese Academy of Medical Sciences were enrolled. Patients involved 17 types of common head and neck cancers, and the prescribed dose covered 5 kinds of dose gradients ranging from 54 Gy to 73.92 Gy. And 1-2 cases per each cancer type (31 cases in total) were randomly selected as the validation set, and the remaining 787 cases were used as the training set to build a deep learning head and neck radiotherapy generalized dose prediction model. Then based on the dose prediction results of this model, a program was written to automatically generate inverse optimization condition scripts, which were sent back to the treatment planning system to achieve AI-assisted radiotherapy plan design. Among the patients who received radiotherapy in our hospital from June 2021 to January 2022, 1 patient for each disease type (17 cases in total) was selected to evaluate the AI-assisted plan design program and evaluate its clinical feasibility using paired t-test. Results:Dose prediction model accuracy evaluation revealed that in the 31-case validation set, there was no statistical difference in the evaluation metrics of clinical concern for organs at risks, except for the D 1 cm3 prediction for spinal cord planning risk volume, which was statistically different compared with the clinical reference plan. The AI-assisted plan design program had higher plan quality metric scores (37.88±6.42) than manual plans (35.00±7.63) in 17 test cases ( t=-1.00, P=0.166). The number of manual adjustments to the inverse optimization conditions was reduced from (5.47±2.97) times to (2.76±1.00) times for the AI-assisted plan compared to the manual-only plan ( t=4.12, P<0.001). And the number of outlined dose shaping structures was reduced from 7.35±3.98 to 3.12±1.18 ( t=5.61, P<0.001). Conclusions:The unified universal model of dose prediction established for different head and neck cancers has high accuracy in dose prediction for all types of head and neck tumor plans. The AI-assisted planning method established in this pattern can reduce the clinical workload of physicists and improve the efficiency of their work.

Artificial intelligence (AI) technologies are being widely applied in radiotherapy. However, the integration of AI into clinical workflows of radiotherapy faces a series of challenges, such as poor model interpretability, domain shifts between clinical application and training data, and the inherent model uncertainties. Therefore, case-specific quality assurance (QA) is essential before deploying AI models in clinical practice. This paper reviews and summarizes QA methodologies for the application of AI models in radiotherapy across four key areas: image registration, image generation, region of interest segmentation, and treatment planning.

Objective:To assess intrafractional errors of the deformation of clinical target volumes (CTVs) during online adaptive radiotherapy (OART) for prostate cancer patients, aiming to provide a basis for online plan optimization.Methods:A retrospective analysis was conducted for 13 prostate cancer patients who received 1.5 T magnetic resonance imaging (MRI)-guided OART (8 Gy × 5 fractions, totaling 65 fractions) at the Department of Radiation Oncology, Cancer Hospital, Chinese Academy of Medical Sciences. The MRI images were collected at the beginning and end of various treatment fractions. Then, the CTVs and organs at risk (OARs) were delineated by the same radiation oncologist. After rigid registration and triangle mesh generation, the surface vertices were extracted. The deformable registration for the CTV surfaces was performed using the thin-plate spline robust point matching (TPS-RPM) algorithm, yielding vertex correspondences. Last, both systematic and random intrafractional errors of CTV deformation were calculated.Results:The average Hausdorff distance (HD) for deformable registration of treatment fractions was (1.68 ± 0.28) mm. The intrafractional systematic errors of CTV deformation were (0.25 ± 3.18) mm (anterior-posterior direction; A-P), (0.89 ± 3.85) mm (left-right direction, L-R), and (-1.98 ± 6.69) mm (superior-inferior direction, S-I). The intrafractional random errors of CTV deformation were determined at (-0.26 ± 1.89) mm (A-P), (-0.08 ± 0.88) mm (L-R), and (-0.04 ± 1.86) mm (S-I).Conclusions:During OART, CTV deformations primarily occur in the S-I direction. Therefore, it is necessary to consider the expanded size of margins in this direction during OART for prostate cancer.

Objective:To construct a general deep learning dose prediction model applicable to radiotherapy for head and neck tumors, establish design methods for artificial intelligence (AI)-assisted radiotherapy plan and evaluate the accuracy of prediction.Methods:Radiotherapy plans of 818 patients who received radiotherapy for head and neck cancers from January 2018 to June 2021 in Cancer Hospital of Chinese Academy of Medical Sciences were enrolled. Patients involved 17 types of common head and neck cancers, and the prescribed dose covered 5 kinds of dose gradients ranging from 54 Gy to 73.92 Gy. And 1-2 cases per each cancer type (31 cases in total) were randomly selected as the validation set, and the remaining 787 cases were used as the training set to build a deep learning head and neck radiotherapy generalized dose prediction model. Then based on the dose prediction results of this model, a program was written to automatically generate inverse optimization condition scripts, which were sent back to the treatment planning system to achieve AI-assisted radiotherapy plan design. Among the patients who received radiotherapy in our hospital from June 2021 to January 2022, 1 patient for each disease type (17 cases in total) was selected to evaluate the AI-assisted plan design program and evaluate its clinical feasibility using paired t-test. Results:Dose prediction model accuracy evaluation revealed that in the 31-case validation set, there was no statistical difference in the evaluation metrics of clinical concern for organs at risks, except for the D 1 cm3 prediction for spinal cord planning risk volume, which was statistically different compared with the clinical reference plan. The AI-assisted plan design program had higher plan quality metric scores (37.88±6.42) than manual plans (35.00±7.63) in 17 test cases ( t=-1.00, P=0.166). The number of manual adjustments to the inverse optimization conditions was reduced from (5.47±2.97) times to (2.76±1.00) times for the AI-assisted plan compared to the manual-only plan ( t=4.12, P<0.001). And the number of outlined dose shaping structures was reduced from 7.35±3.98 to 3.12±1.18 ( t=5.61, P<0.001). Conclusions:The unified universal model of dose prediction established for different head and neck cancers has high accuracy in dose prediction for all types of head and neck tumor plans. The AI-assisted planning method established in this pattern can reduce the clinical workload of physicists and improve the efficiency of their work.

Artificial intelligence (AI) technologies are being widely applied in radiotherapy. However, the integration of AI into clinical workflows of radiotherapy faces a series of challenges, such as poor model interpretability, domain shifts between clinical application and training data, and the inherent model uncertainties. Therefore, case-specific quality assurance (QA) is essential before deploying AI models in clinical practice. This paper reviews and summarizes QA methodologies for the application of AI models in radiotherapy across four key areas: image registration, image generation, region of interest segmentation, and treatment planning.

Objective:To establish a radiotherapy treatment planning process of high ventilation functional lung avoided (HVFLA) for thoracic tumors based on 4D-CT lung ventilation functional images and determine the treatment planning strategy of HVFLA radiotherapy, and so as to provide support for the clinical trials of HVFLA radiotherapy in thoracic cancer patients.Methods:A deep learning-based 4D-CT lung ventilation functional imaging model was established and integrated into the radiotherapy treatment planning process. Furthermore, ten thoracic cancer patients with 4D-CT simulation positioning were retrospectively enrolled in this study. The established model was used to obtain the 4D-CT lung ventilation functional imaging for each patient. According to the relative value of lung ventilation, the lung ventilation areas are equally segmented into high, medium and low lung ventilation and then imported them into Pinnacle 3 treatment planning system. According to the prescription dose of target and dose constraints of organ at risks (OARs), the clinical and HVFLA treatment plans were designed for each patient using volumetric modulated radiotherapy technique, and each plan should meet the clinical requirements and adding dose constraints of high ventilation functional lung for HVFLA plan. The dosimetric indexes of the target, OARs (lungs, heart and cord) and high functional lung (HFL) were used to evaluated the plan quality. The dosimetric indexes included D2, D98 and mean dose of target, V5, V10, V20, V30 and mean dose of lungs and HFL, V30, V40 and mean dose of heart, and D1 cm 3 of cord. Paired samples t-test was used for statistical analysis of the two groups of plans. Results:The target and OARs of the clinical plan and HVFLA plan meet the clinical requirements. The HVFLA plan resulted in a statistically significant reduction in the mean dose, V5, V10, V20, and V30 of the high functional lung by 1.2 Gy, 5.9%, 4.2%, 2.6%, and 2.3%, respectively ( t=-8.07, 4.02, -6.02, -7.06, -6.77, P<0.05). There was no statistical difference in the dosimetric indexes of lungs, heart and cord. Conclusions:We established the treatment planning process of HVFLA radiotherapy based on 4D-CT lung ventilation functional images. The HVFLA plan can effectively reduce the dose of HFL, while the doses of lungs, heart and cord had no significant difference compared with the clinical plan. The strategy of HVFLA radiotherapy planning is feasible to provide support for the implementation of HVFLA radiotherapy in thoracic cancer patients.

Objective:To verify the feasibility of using Elekta accelerated go live (AGL) standard process for the acceptance of multiple accelerators.Methods:The beams of three accelerators were adjusted by PTW Beamscan three-dimensional water tank to reach the AGL standard. Dose verification was performed for three accelerators that met AGL standards. A simple field test example from Cancer Hospital Chinese Academy of Medical Sciences was used to compare the MapCheck 3 surface dose measurement results with the surface dose calculated by the same accelerator model. Images of 10 patients including head and neck, esophagus, breast, lung and rectum were randomly selected. volumetric-modulated arc therapy (VMAT) and intensity modulated radiation therapy (IMRT) treatment techniques were used for planning design, and the measured dose of ArcCheck was compared with the planned dose calculated by the same accelerator model. One-way ANOVA was used to statistically analyze the passing rates of two-dimensional and three-dimensional dose verification.Results:The 6 MV X-ray percentage depth dose at 10 cm underwater (PDD 10) of three accelerators was 67.45%, 67.36%, 67.47%, and the maximum deviation between the three accelerators was 0.11%. The 6 MV flattenting filter free (FFF) mode X-ray PDD 10 was 67.33%, 67.20%, 67.20%, and the maximum deviation between the three accelerators was 0.13%. All required discrete point doses on each energy 30 cm×30 cm Profile spindle of the three accelerator X-rays deviated less than ±1% from the standard data. Absolute γ analysis was performed on the results of MapCheck 3 two-dimensional dose matrix validation. Under the 10% threshold of 2 mm/3% standard, the average passing rate of the test cases in Cancer Hospital Chinese Academy of Medical Sciences was above 99%, and the difference was not statistically significant ( P>0.05). Absolute γ analysis was performed on the ArcCheck verification results. Under the 10% threshold, the pass rate of 2 mm/3% was all above 95%, the maximum average passing rate of the three accelerators with different energy and different treatment techniques was 0.28% (6 MV, VMAT), 0.19%(6 MV FFF, VMAT), 0.56% (6 MV, IMRT) and 0.05% (6 MV FFF, IMRT), and the difference was not statistically significant ( P>0.05). Conclusion:Compared with traditional accelerator acceptance process, the acceptance time of each accelerator is shortened by 4-6 weeks by using the AGL standard process, and the radiotherapy plan of patients can be interchangeably executed among different accelerators.

Objective:To develop a deep learning method to predict the anatomical images of nasopharyngeal carcinoma patients during the treatment course, which could detect the anatomical variation for specific patients in advance.Methods:Imaging data including planning CT (pCT) and cone-beam CT (CBCT) for each fraction of 230 patients with T 3-T 4 staging nasopharyngeal carcinoma who treated in Cancer Hospital Chinese Academy of Medical Sciences from January 1, 2020 to December 31, 2022 were collected. The anatomical images of week k+1 were predicted using a 3D Unet model with inputs of pCT, CBCT on days 1-3, and CBCT of weeks 2- k. In this experiment, we trained four models to predict anatomical images of weeks 3-6, respectively. The nasopharynx gross tumor volume (GTV nx) and bilateral parotid glands were delineated on the predicted and real images (ground truth). The performance of models was evaluated by the consistence of the delineation between the predicted and ground truth images. Results:The proposed method could predict the anatomical images over the radiotherapy course. The contours of interest in the predicted image were consistent with those in the real image, with Dice similarity coefficient of 0.96, 0.90, 0.92, mean Hausdorff distance of 3.28, 4.18 and 3.86 mm, and mean distance to agreement of 0.37, 0.70, and 0.60 mm, for GTV nx, left parotid, and right parotid, respectively. Conclusion:This deep learning method is an accurate and feasible tool for predicting the patient's anatomical images, which contributes to predicting and preparing treatment strategy in advance and achieving individualized treatment.

Objective To explore the effect of auto-segmentation based on deep learning(DL)and Atlas during online adaptive MRI-guided radiotherapy.Methods Totally 15 prostate cancer patients undergoing MRI-guided online adaptive radiotherapy at some hospital from January 2020 to September 2021 were selected and divided into a training set(12 cases)and a test set(3 cases)by random sampling method.With the training set data the models of clinical target volume(CTV)and organs at risk(OAR)by DL and Atlas segmentation were established,and with the test set data the two segmentation models were modified and the modification lengths were recorded.DL and Atlas segmentation methods were compared on segmentation efficiency and accuracy in terms of Dice similarity coefficient(DSC),Hausdorff distance(HD)and mean distance to agreement(MDA).A joint auto-segmentation scheme based on combined DL and Atlas was constructed with considerations on the advantages and characteristics of the two methods,which was compared with the schemes respectively based on DL or Atlas from the aspect of the time consumed for segmentation.Results Accuracy comparison showed Atlas segmentation model behaved better significantly than DL model for CTV(P＜0.05),while obviously worse than the latter for DSC and MDA in bladder and rectum(P＜0.05).The doctor took 9.4 min in average for CTV and OAR modification based on DL model and 12 min in average for Atlas-model-based modification.The joint auto-segmentation scheme only needed 8 min in average for CTV and OAR modification,which gained advantages over the schemes based on DL or Atlas.Conclusion The auto-segmentation based on combined DL and Atlas during online adaptive MRI-guided radiotherapy behaves well in low time consumption,high accuracy and efficiency.[Chinese Medical Equipment Journal,2024,45(6):59-64]