研究目的：

       在线自适应放疗（ART）通过在放疗过程中获取患者治疗当次的图像数据，实时调整放疗计划，从而在盆腔肿瘤的放疗中具有显著优势。然而，目前该项新技术在宫颈癌中的应用研究较少。本研究旨在通过三部分临床研究实现以下目的：1.探讨每日基于迭代锥形束CT (iCBCT)引导的在线ART在宫颈癌术后中的最佳计划靶区（PTV）外放边界（margin），及缩小边界带来的剂量学、临床获益。2.开展基于iCBCT引导的在线ART在宫颈癌放疗中的前瞻性I期临床研究，评估患者初步疗效及急性毒副反应。3.分析基于iCBCT引导的在线ART在宫颈癌中自动化轮廓的精确度和剂量学结果，探索培训在线自适应师时需重点关注的区域。

研究方法：

第一部分：基于iCBCT引导的在线自适应放疗在宫颈癌术后PTV-CTV外放边界研究

       125个宫颈癌/子宫内膜癌术后在线ART分次作为训练集，在分次放疗后扫描的iCBCT₃上再次勾画临床靶区（CTV）。将用以产生在线自适应计划的放疗前iCBCT₁中CTV₁进行5、7、10、12、15mm的均匀外放生成相应PTV₁，评估包绕CTV₃所需的最小PTV₁边界。253个在线ART分次被用于结果验证，并探讨进一步缩小PTV边界的可能。使用既往接受过iCBCT图像验证的125个放疗分次模拟自适应计划，分别采用最小自适应PTV边界和传统PTV外放边界（7-15mm）生成两套自适应计划，并进行剂量学比较。

第二部分：在线自适应放疗在宫颈癌中的疗效及急性毒副反应研究：一项前瞻性单臂I期临床研究

        前瞻性入组宫颈癌/子宫内膜癌术后、宫颈癌根治性在线ART的患者。术后在线ART计划采用均匀外放5mm的PTV边界，45Gy/25次或50.4Gy/28次，根治性在线ART采用子宫体CTV-U外放10mm、其余CTV外放5mm的PTV边界，50.4Gy/28次。每分次的在线ART生成adapted计划和scheduled计划。在线ART期间、内照射期间、治疗后1个月分别评估急性毒副反应(放疗开始后90天内)，按照常见不良事件评价标准（CTCAE）5.0分级。患者报道的毒副反应来自EORTC QLQ-C30问卷。

第三部分：在线自适应师培训的探索性研究：自动化轮廓评估与剂量学研究

       125个分次的宫颈癌术后在线ART，140个分次的宫颈癌根治性在线ART被纳入分析。经医师手动修改的靶区作为标准靶区轮廓，生成的计划定义为S-ART（Supervised ART）。每分次未手动修改的自动轮廓经模拟生成的自适应计划定义为自动化在线ART(Automatic ART,A-ART)。CTV被细分为16个独立区域，通过平均表面距离（ASD）、质心偏差（DC）、Dice相似系数（DSC）和95% 豪斯多夫距离（95% HD）评估自动勾画CTV与手动修改CTV之间的相似程度。重新计算了A-ART在手动修改后的靶区及危及器官轮廓上的剂量，并与S-ART剂量进行比较。

研究结果：

第一部分： 从iCBCT₁扫描开始至治疗后iCBCT₃扫描结束，平均耗时约22分钟。均匀外放5mm的PTV-N₁可以覆盖训练集中100%的淋巴引流区CTV-N₃（125/125分次，p＞0.05）和98%的阴道断端CTV-V₃（123/125分次，p＞0.05），验证集中100%的CTV-N₃（253/253分次，p＞0.05）和99%的CTV-V₃（250/253分次，p＞0.05）。在满足CTV覆盖率≥95%的情况下，PTV-N边界可以缩小至4mm（p＞0.05）。与传统7—15mm外放边界的计划比，采用5mm边界的adapted计划显著降低了包括膀胱、直肠、小肠、骨髓、股骨头在内的危及器官受照剂量。

第二部分：17例术后患者共完成434个分次的在线ART, 27例根治性患者完成756个分次的在线ART。所有分次均选择了adapted计划。与scheduled计划相比，adapted计划显著改善了CTV和PTV的剂量覆盖，显著降低直肠、小肠、骨髓、股骨头的受照剂量，此外，在宫颈癌术后在线ART中，adapted计划还降低了膀胱受照剂量。术后在线ART患者在治疗后9个月随访时没有发生局部复发或远处转移,2级胃肠道急性反应为24%，没有出现泌尿系统毒副反应。在根治性在线ART中，入组患者治疗后1月疗效评估示肿瘤完全缓解，1、2、3级的胃肠道、泌尿系统急性毒副反应分别为26%，19%，4%，和7%，0，0。在患者报道的毒副反应方面，在线ART期间在功能领域和总体健康状况领域得分显著降低，在症状领域得分显著升高，在近距离治疗期间开始改善，治疗后1个月基本恢复到基线水平。

第三部分：A-ART和S-ART的CTV具有较高的重合度，各区域的DSC均值大于0.75。自动化CTV-U轮廓的相似性较比其他区域相对较大，ASD、DC、DSC和95% HD分别为2.67±1.79mm、17.17±12mm、0.76±0.12和10.45±5.68mm。在术后在线ART的CTV-V和根治性在线ART的CTV-C中，下半区域的自动化精度要优于上半区域。在共计265例CTV-N中，下部CTV的一致性差异相对较大。术后A-ART的adapted计划中CTV的V_100%均＞97%，PTV的V_100%均＞91%。根治性A-ART的adapted计划中CTV的V_100%均＞99%，PTV的V_100%均＞93%。

研究结论：

       1. 在宫颈癌术后放疗中，在线ART能将PTV外放边界缩小至5mm。在满足靶区充分剂量覆盖的同时，直肠和膀胱等危及器官的剂量学可明显改善。

       2. 基于缩小后的PTV外放边界，采用每日在线ART在宫颈癌中取得理想的初步临床疗效、更好的靶区剂量覆盖、降低周围正常组织的受照剂量，以及减少急性毒副反应。这尚需在更大的随机对照队列中进行前瞻性的验证。

       3. 基于iCBCT的在线自适应平台利用卷积神经网络人工智能算法生成的自动化轮廓及相应的自适应计划，在宫颈癌术后和根治性放疗中是临床可接受的。子宫体、宫颈、上段阴道残端和低位淋巴引流区是培训自适应师时需要重点关注的区域。

Objectives:

       Online adaptive radiation therapy (ART) has significant advantages in the radiation therapy of pelvic tumors by acquiring the patients' image data during the treatment process and adjusting the radiation therapy plan in real time. However, there are currently few studies on the application of this new technology in cervical cancer. The following objectives were achieved by conducting a three-part study: 1. To determine the optimal planning target volume (PTV) margin for postoperative cervical cancer by daily iterative cone beam computed tomography (iCBCT) guided online ART, and the benefit of reducing PTV margins. 2. To investigate the details and workflow of applying iCBCT-Linac in cervical cancer radiation therapy, and evaluate the efficacy and acute toxicity of patients. 3. To evaluate the accuracy and dosimetric outcomes of automatic contouring of iCBCT-based online ART in cervical cancer, and explore areas that need to be focused when training online adapter.

Methods:

Part I: Evaluation of PTV margins with daily iCBCT guided online adaptive radiation therapy for postoperative treatment of cervical cancer

       A total of 125 fractions from postoperative cervical and endometrial cancer treated with daily iCBCT-guided online ART were regarded as training sets, and the clinical target volumes (CTV) were contoured from post-treatment iCBCT₃ scans. Uniform three-dimensional planning margins of 5, 7, 10, 12, and 15 mm were added to the pre-treatment CTV₁ to generate PTV₁, to assess the minimum planning margins required to encompass the CTV₃. The minimum PTV margin was verified on a validation sets of 253 fractions, and further margin reduction were studied. A separate cohort of 125 iCBCT scans previously used for daily position verification were emulated for daily adaptive replanning, and two reference plans on the same fraction were generated and compared, adopting reduced margins and conventional margin (7-15mm).

Part II: Efficacy and acute toxicity of online adaptive radiation therapy in cervical cancer: a prospective single-arm Phase I clinical study

       Postoperative cervical and endometrial cancer patients and uterine cervical cancer patients were prospectively enrolled. The CTV of postoperative cervical cancer was expanded by a uniform three-dimensional planning margin of 5 mm to generate the PTV, and PTV received 45 Gy or 50.4 Gy in 1.8 Gy daily fraction. A uniform 10 mm margin was used to cover more variable uterus (PTV-U), and 5 mm margin was used for other PTV in uterine cervical cancer, and PTV received 50.4 Gy in 1.8 Gy daily fraction. An adapted plan and a scheduled plan were generated each online ART fraction. Acute toxicities (within 90 days after initiation of radiotherapy) were assessed before treatment, weekly during treatment, and 1 month after treatment, and graded according to the CTCAE 5.0. Patient-reported acute toxicity was derived from the EORTC Quality of Life Questionnaire (QLQ)-C30 questionnaire.

Part III: An exploratory study of online adapter training: auto-contouring and dosimetric evaluation

       A total of 125 online ART fractions from postoperative cervical cancer and 140 online ART fractions from uterine cervical cancer treated with daily iCBCT-guided online ART were analyzed. The contouring of target volume manually corrected by physicians was regarded as the standard contouring, and generated online ART was defined as Supervised online ART (S-ART). The Automatic online ART (A-ART) was generated using the fully automatic workflow without manual corrections. The CTV was subdivided into 16 separate regions, and the average surface distance dice (ASD), deviation of centroid (DC), dice similarity coefficient (DSC), and 95% Hausdorff distance (95% HD) were used to evaluate contouring for the above portions. The dosimetric outcomes of A-ART were re-calculated with manually edited contours. Dosimetric results from A-ART plans were compared to S-ART plans, to evaluate target volumes and organs at risk dose changes.

Results:

Part I: The average time from pre-treatment iCBCT₁ to post-treatment iCBCT₃ was proximately 22 min. A uniform PTV margin of 5 mm could encompass nodal CTV₃ (CTV-N) in 100% of the fractions (175/175, p>0.05) and vaginal CTV₃ (CTV-V) in 98% of the fractions (172/175, p>0.05). The 5 mm PTV margin could cover 100% of CTV-N₃ (253/253, p>0.05) and 99% of CTV-V₃ (250/253, p>0.05) in validation sets. The PTV-N margin could be further reduced to 4 mm if ≥ 95% CTV coverage was predicted to be achieved. The adapted plan with a 5 mm margin significantly improved organs at risk dosimetry compared with the conventional margin (7-15mm) plan.

Part II: A total of 125 online ART fractions from 17 enrolled postoperative cervical cancer patients, and 756 online ART fractions from 27 uterine cervical cancer patients were analyzed. The adapted plan was selected in all fractions. Compared with scheduled plan, adapted plan significantly improved the dosimetry coverage for CTV and PTV, and significantly improved the rectum, bowel, bone marrow, femur head left and right dosimetry. In addition, the bladder dosimetry was improved in the postoperative adapted plan. No local recurrence or distant metastasis have occurred 9 months after the completion of treatment in postoperative cervical cancer. Grade 2 acute gastrointestinal was 24%, and no acute urinary toxicities was occurred. For uterine cervical cancer patients, a comprehensive review one month after treatment showed that efficacy evaluation was complete response, and grade 1, 2, and 3 gastrointestinal and urinary system toxicities were 26%, 19%, 4%, and 7%, 0, 0, respectively. Mean scores in the functioning scales significantly decreased and mean scores in the symptom scales significantly increased after online ART initiation. Improvement began during the brachytherapy period and basically returned to the baseline level at 1 month after treatment.

Part III: The A-ART and S-ART had high overlap rates, with an average DSC value greater than 0.75. The automatic CTV-U had the relatively largest consistency differences, with ASD, DC, and 95% HD being 2.67 ± 1.79 mm, 17.17 ± 12 mm, and 10.45 ± 5.68 mm, respectively. For CTV-V in the postoperative group and CTV-C in the radical group, the automation accuracy of the lower region was better than that in the upper region. For 265 paired CTV- N, the consistency differences of lower CTV-N_left and lower CTV- N_right were relatively large. In adapted plan of postoperative A-ART, the V_100% of CTV was >97%, and the V_100% of PTV was > 91%. In adapted plan of radical A-ART, the V_100% of CTV was > 99%, and the V_100% of PTV was > 93%.

Conclusions:

       1. In the postoperative treatment of cervical cancer, online ART could reduce PTV margin to 5 mm. The daily online adapted plans significantly decreased the dose to organs at risk, while maintaining a therapeutic dose to the target volume.

       2. On the basis of using the reduced PTV margin, daily online ART achieved ideal clinical efficacy, better target dosimetry coverage, surrounding normal tissue dosimetry benefits and lower acute toxicity. These need to be prospectively verified in a larger randomized controlled cohort.

       3. The auto-contouring and dosimetric outcomes from the online ART adaptive platform by the AI algorithm of using convolutional neural network were considered clinically acceptable for cervical cancer. The uterus, cervix, upper vaginal cuff, and lower nodal CTV were the regions that need to be focused on in training "adapter".