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 evaluate the role of NOD-like receptor protein 3 (NLRP3) inflammasome-mediated microglia activation in myocardial ischaemia-reperfusion-induced brain injury in mice.Methods:Fifty-two SPF healthy male wild-type C57BL/6 mice and 52 NLRP3 -/- mice, aged 8-10 weeks, were divided into 4 groups ( n=26 each) using a random number table method: wild type sham operation group (W-S group), wild type myocardial ischemia-reperfusion group (W-IR group), NLRP3 -/- sham operation group (NLRP3 -/--S group), and NLRP3 -/- myocardial ischemia-reperfusion group (NLRP3 -/--IR group). The myocardial ischemia-reperfusion-induced brain injury model was established by ligating the left anterior descending coronary artery for 45 min followed by 24 h of reperfusion in anesthetized mice. The cognitive function was evaluated using the modified Morris water maze test at 24 h of reperfusion. The mice were sacrificed after blood specimens were collected, and brain tissues were obtained for measurement of the blood-brain barrier permeability and water content, for microscopic examination of the pathological changes of brain tissues, and for determination of serum S-100β protein and neuron-specific enolase (NSE) concentrations, contents of interleukin-1 beta (IL-1β), IL-6 and tumor necrosis factor-alpha (TNF-α) in hippocampal tissues (by enzyme-linked immunosorbent assay), expression of NLRP3, apoptosis-associated speck-like protein (ASC), cleaved cysteine aspartate protease 1 (cleaved-caspase-1), gasdermin D (GSDMD), ionized calcium-binding adapter molecule 1 (Iba-1), and occludin in hippocampal tissues (by immunofluorescence and/or Western blot). The apoptosis rate of neurons and density of dendritic spine were calculated. Results:Compared with sham operation group, the escape latency was significantly prolonged, the number of crossing the original platform was decreased, and the time spent in the target quadrant was shortened, the concentrations of serum S-100β protein and NSE were increased, the blood-brain barrier permeability and brain water content were increased, the dendritic spine density in the hippocampal CA1 area was decreased, the contents of IL-1β, IL-6 and TNF-α were increased, the expression of NLRP3, ASC, cleaved-caspase-1, GSDMD and Iba-1 was up-regulated, and the expression of occludin was down-regulated ( P<0.05), and the pathological injury to brain tissues was found in ischemia-reperfusion group. Compared with W-IR group, the escape latency was significantly shortened, the number of crossing the original platform was increased, and the time spent in the target quadrant was prolonged, the concentrations of serum S-100β protein and NSE were decreased, the blood-brain barrier permeability and brain water content were decreased, the dendritic spine density in the hippocampal CA1 area was increased, the contents of IL-1β, IL-6 and TNF-α were decreased, the expression of NLRP3, ASC, cleaved-caspase-1, GSDMD and Iba-1 was down-regulated, and the expression of occludin was up-regulated ( P<0.05), and the pathological injury to brain tissues was alleviated in NLRP3 -/--IR group. Conclusions:NLRP3 inflammasome-mediated microglia activation is involved in myocardial ischaemia-reperfusion-induced brain injury in mice.

Objective:To analyze the learning curve of TiRobot-assisted lumbar pedicle screw fixation (LPSF) by cumulative sum (CUSUM) test method.Methods:The clinical data of 50 patients who underwent TiRobot-assisted LPSF from January 2020 to December 2022 in Beijing Friendship Hospital, Capital Medical University were retrospectively analyzed. CUSUM analysis and learning curve fitting were performed with robot usage time as the main indicator with the time for each step refined (robot registration time, path planning time and guide wire placement time), to select the best learning curve fitting model with the R2 value closest to 1. Using the turning point of the learning curve as the boundary, the learning curve was divided into two stages as learning stage and maturity stage, and then the observation indexes were compared between the two stages. Results:All 50 patients successfully completed the surgery without perioperative complications, with a total of 244 pedicle screws implanted. The total robot usage time and robot registration time showed a gradually decreasing trend with the increase of case number, and the learning curves were successfully fitted and reached their peaks at the seventeenth and thirteenth cases respectively. The entire learning process was divided into learning stage (17 cases) and maturity stage (33 cases) based on the turning point of the learning curve of total robot usage time. The path planning time and guide wire placement time did not show significant changes with the increase in the case number. The total robot usage time, robot registration time and the intraoperative blood loss in the learning stage were significantly higher than those in the maturity stage: (35.35 ± 1.58) min vs. (30.61 ± 0.43) min, (20.83 ± 1.56) min vs. (14.94 ± 0.29) min and 400 (150, 500) ml vs. 200 (110, 300) ml, the guide wire placement time of per screw was significantly lower than that in the maturity stage: 2.00 (1.83, 2.34) min/screw vs. 2.33 (2.13, 2.69) min/screw, and there were statistical differences ( P<0.05 or <0.01). There were no statistical difference in the path planning time, path planning time of per screw, guide wire placement time and the accuracy of screw placement between two stages ( P>0.05). Conclusions:TiRobot-assisted LPSF is a new technology with safety and effectiveness, and it has a relatively short learning curve. To achieve technological maturity, at least 17 surgeries are required with accumulated experience, and the robot registration is the main step of the learning process. After reaching maturity stage, the robot usage time is significantly shortened and intraoperative trauma is significantly reduced while the relatively high screw placement accuracy is ensured.

Objective:To evaluate the role of NOD-like receptor protein 3 (NLRP3) inflammasome-mediated microglia activation in myocardial ischaemia-reperfusion-induced brain injury in mice.Methods:Fifty-two SPF healthy male wild-type C57BL/6 mice and 52 NLRP3 -/- mice, aged 8-10 weeks, were divided into 4 groups ( n=26 each) using a random number table method: wild type sham operation group (W-S group), wild type myocardial ischemia-reperfusion group (W-IR group), NLRP3 -/- sham operation group (NLRP3 -/--S group), and NLRP3 -/- myocardial ischemia-reperfusion group (NLRP3 -/--IR group). The myocardial ischemia-reperfusion-induced brain injury model was established by ligating the left anterior descending coronary artery for 45 min followed by 24 h of reperfusion in anesthetized mice. The cognitive function was evaluated using the modified Morris water maze test at 24 h of reperfusion. The mice were sacrificed after blood specimens were collected, and brain tissues were obtained for measurement of the blood-brain barrier permeability and water content, for microscopic examination of the pathological changes of brain tissues, and for determination of serum S-100β protein and neuron-specific enolase (NSE) concentrations, contents of interleukin-1 beta (IL-1β), IL-6 and tumor necrosis factor-alpha (TNF-α) in hippocampal tissues (by enzyme-linked immunosorbent assay), expression of NLRP3, apoptosis-associated speck-like protein (ASC), cleaved cysteine aspartate protease 1 (cleaved-caspase-1), gasdermin D (GSDMD), ionized calcium-binding adapter molecule 1 (Iba-1), and occludin in hippocampal tissues (by immunofluorescence and/or Western blot). The apoptosis rate of neurons and density of dendritic spine were calculated. Results:Compared with sham operation group, the escape latency was significantly prolonged, the number of crossing the original platform was decreased, and the time spent in the target quadrant was shortened, the concentrations of serum S-100β protein and NSE were increased, the blood-brain barrier permeability and brain water content were increased, the dendritic spine density in the hippocampal CA1 area was decreased, the contents of IL-1β, IL-6 and TNF-α were increased, the expression of NLRP3, ASC, cleaved-caspase-1, GSDMD and Iba-1 was up-regulated, and the expression of occludin was down-regulated ( P<0.05), and the pathological injury to brain tissues was found in ischemia-reperfusion group. Compared with W-IR group, the escape latency was significantly shortened, the number of crossing the original platform was increased, and the time spent in the target quadrant was prolonged, the concentrations of serum S-100β protein and NSE were decreased, the blood-brain barrier permeability and brain water content were decreased, the dendritic spine density in the hippocampal CA1 area was increased, the contents of IL-1β, IL-6 and TNF-α were decreased, the expression of NLRP3, ASC, cleaved-caspase-1, GSDMD and Iba-1 was down-regulated, and the expression of occludin was up-regulated ( P<0.05), and the pathological injury to brain tissues was alleviated in NLRP3 -/--IR group. Conclusions:NLRP3 inflammasome-mediated microglia activation is involved in myocardial ischaemia-reperfusion-induced brain injury in mice.

Objective:To analyze the learning curve of TiRobot-assisted lumbar pedicle screw fixation (LPSF) by cumulative sum (CUSUM) test method.Methods:The clinical data of 50 patients who underwent TiRobot-assisted LPSF from January 2020 to December 2022 in Beijing Friendship Hospital, Capital Medical University were retrospectively analyzed. CUSUM analysis and learning curve fitting were performed with robot usage time as the main indicator with the time for each step refined (robot registration time, path planning time and guide wire placement time), to select the best learning curve fitting model with the R2 value closest to 1. Using the turning point of the learning curve as the boundary, the learning curve was divided into two stages as learning stage and maturity stage, and then the observation indexes were compared between the two stages. Results:All 50 patients successfully completed the surgery without perioperative complications, with a total of 244 pedicle screws implanted. The total robot usage time and robot registration time showed a gradually decreasing trend with the increase of case number, and the learning curves were successfully fitted and reached their peaks at the seventeenth and thirteenth cases respectively. The entire learning process was divided into learning stage (17 cases) and maturity stage (33 cases) based on the turning point of the learning curve of total robot usage time. The path planning time and guide wire placement time did not show significant changes with the increase in the case number. The total robot usage time, robot registration time and the intraoperative blood loss in the learning stage were significantly higher than those in the maturity stage: (35.35 ± 1.58) min vs. (30.61 ± 0.43) min, (20.83 ± 1.56) min vs. (14.94 ± 0.29) min and 400 (150, 500) ml vs. 200 (110, 300) ml, the guide wire placement time of per screw was significantly lower than that in the maturity stage: 2.00 (1.83, 2.34) min/screw vs. 2.33 (2.13, 2.69) min/screw, and there were statistical differences ( P<0.05 or <0.01). There were no statistical difference in the path planning time, path planning time of per screw, guide wire placement time and the accuracy of screw placement between two stages ( P>0.05). Conclusions:TiRobot-assisted LPSF is a new technology with safety and effectiveness, and it has a relatively short learning curve. To achieve technological maturity, at least 17 surgeries are required with accumulated experience, and the robot registration is the main step of the learning process. After reaching maturity stage, the robot usage time is significantly shortened and intraoperative trauma is significantly reduced while the relatively high screw placement accuracy is ensured.

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 improve the performance of auto-segmentation of prostate target area and organs-at-risk in online magnetic resonance image and enhance the efficiency of magnetic resonance imaging-guided adaptive radiotherapy(MRIgART)for prostate cancer.Methods A retrospective study was conducted on 40 patients who underwent MRIgART for prostate cancer,including 25 in the training set,5 in the validation set,and 10 in the test set.The planning CT images and corresponding contours,along with online MR images,were registered and input into a deep learning network for online MR image auto-segmentation.The proposed method was compared with deformable image registration(DIR)method and single-MR-input deep learning(SIDL)method.Results The overall accuracy of the proposed method for auto-segmentation was superior to those of DIR and SIDL methods,with average Dice similarity coefficients of 0.896 for clinical target volume,0.941 for bladder,0.840 for rectum,0.943 for left femoral head and 0.940 for right femoral head,respectively.Conclusion The proposed method can effectively improve the accuracy and efficiency of auto-segmentation in MRIgART for prostate cancer.

Objective To evaluate the effects of the multiple factors especially image geometric accuracy of the imaging system on the segmentations of target areas and organs-at-risk.Methods The study used phantoms to test the imaging performance of the 1.5T magnetic resonance(MR)linear accelerator system,including the assessments of MR image geometric distortion and the segmentation errors caused by factors such as image geometric distortion.Model 604-GS large field MR image distortion phantom was used to explore the geometric distortion of the MR images for MR-guided radiotherapy;and CIRS Model 008z upper abdominal phantom was used to analyze the segmentation errors of target areas and organs-at-risk.Results The average geometric distortion and maximum distortion of 3D T1WI-FFE images vs 3D T2WI-TSE images were 0.54 mm vs 0.53 mm and 1.96 mm vs 1.68 mm,respectively;and the control points of the large distortions were distributed at the edges of the phantom,which was consistent with the MR imaging characteristics previously reported.Compared with CT-based segmentation contour,the MDA was 1.17 mm and DSC was 0.91 for 3D T1WI-FFE,while MDA was 0.86 mm and DSC was 0.94 for 3D T2WI-TSE.Conclusion The study quantitatively assesses the geometric accuracy of the imaging system for MR-guided radiotherapy.The phantom-based contour analysis reveals that with CT image as gold standard,the segmentation error in MRI images meets the clinical requirements,and that 3D T2WI-TSE image is advantageous over 3D T1WI-FFE image in segmentation accuracy.

Objective:To evaluate the feasibility of using MRI-only simulation images for dose calculation of both photon and proton radiotherapy for nasopharyngeal carcinoma cases.Methods:T 1-weighted MRI images and CT images of 100 patients with nasopharyngeal carcinoma treated with radiotherapy in Cancer Hospital of Chinese Academy of Medical Sciences from January 2020 to December 2021 were retrospectively analyzed. MRI images were converted to generate pseudo-CT images by using deep learning network models. The training set, validation set and test set included 70 cases, 10 cases and 20 cases, respectively. Convolutional neural network (CNN) and cycle-consistent generative adversarial neural network (CycleGAN) were exploited. Quantitative assessment of image quality was conducted by using mean absolute error (MAE) and structural similarity (SSIM), etc. Dose assessment was performed by using 3D-gamma pass rate and dose-volume histogram (DVH). The quality of pseudo-CT images generated was statistically analyzed by Wilcoxon signed-rank test. Results:The MAE of the CNN and CycleGAN was (91.99±19.98) HU and (108.30±20.54) HU, and the SSIM was 0.97±0.01 and 0.96±0.01, respectively. In terms of dosimetry, the accuracy of pseudo-CT for photon dose calculation was higher than that of the proton plan. For CNN, the gamma pass rate (3 mm/3%) of the photon radiotherapy plan was 99.90%±0.13%. For CycleGAN, the value was 99.87%±0.34%. The gamma pass rates of proton radiotherapy plans were 98.65%±0.64% (CNN, 3 mm/3%) and 97.69%±0.86% (CycleGAN, 3 mm/3%). For DVH, the dose calculation accuracy in the photon plan of pseudo-CT was better than that of the proton plan.Conclusions:The deep learning-based model generated accurate pseudo-CT images from MR images. Most dosimetric differences were within clinically acceptable criteria for photon and proton radiotherapy, demonstrating the feasibility of an MRI-only workflow for radiotherapy of nasopharyngeal cancer. However, compared with the raw CT images, the error of the CT value in the nasal cavity of the pseudo-CT images was relatively large and special attention should be paid during clinical application.

Objective To observe the effect of ischemia postconditioning(IPostC)on myocardial ischemia-reperfusion(I/R)in-jury in atherosclerosis(Apoe-/-+AS)model mice,and to study the mechanism of the effect of AAV9-HIF-1α on IPostC.Methods Sixty Apoe-/-+AS model mice were randomly divided into 6groups:control group(AS-sham group),Apoe-/-+AS-myocardial ischemia-reperfusion group(AS-I/R group),Apoe-/-+AS-ischemia postconditioning group(AS-IPostC group),Apoe-/-+AS-AAV9-HIF-1 α+ischemia postconditioning group(AS-AAV9-HIF-1 α+IPostC group),Apoe-/-+AS-AAV9-NC+ischemia postconditioning group(AS-AAV9-NC+IPostC group),Apoe-/-+AS-AAV9-HIF-1α+HIF-1α in-hibitor+ischemia postconditioning group(AS-AAV9-HIF-1α+2ME2+IPostC group),6mice in each group.Establish myocardial I/R models and IPostC models;adeno-associated virus-9 with human cytomegalovirus promoter negative control(AAV9-NC)and ad-eno-associated virus-9 with human cytomegalovirus promoter expressing HIF-1α(AAV9-HIF-1α)was separately injected through the tail vein.The histopathological changes of myocardium was observed,reactive oxygen species(ROS),adenosine triphosphate(ATP),and the activities of mitochondrial respiratory enzyme were detected;hypoxia inducible factor-1 α(HIF-lα),ATP synthase subunit b(ATP5F1),adenylate transporter(ANT)were detected by Western blot assay.Results Compared with AS-sham group,myocardial tissue injury was serious,ROS content was increased,ATP content was significantly decreased in AS-I/R group;compared with the AS-I/R group,ROS content was decreased,ATP content was increased in the AS-IPostC group,but mitochondrial respiratory enzyme activity was still lower,while myocardial injury was significantly decreased in the AS-AAV9-HIF-1α+IPostC group,and mi-tochondrial respiratory function was significantly improved.Conclusion During myocardial I/R injury in atherosclerotic mice,oxygen free radical activated by oxidative stress was released,mitochondrial respiratory function was weakened,cell energy was decreased,and myocardial cell was damaged,while AAV9-HIF-1α+IPostC can enhance myocardial mitochondrial respiratory function to a certain ex-tent through the above mechanisms.