AccuLearning系统在宫颈癌放疗中自动勾画性能评估

Performance evaluation of the AccuLearning system for automated delineation in cervical cancer radiotherapy

摘要

目的：评估AccuLearning系统在宫颈癌放疗中自动勾画的性能。方法：回顾性纳入兰州大学第一医院2022年5月~2025年5月接受放疗的120名宫颈癌患者定位CT影像及手动勾画结构。随机选取80例作为训练集，用于开发和训练自动勾画模型，并对剩余40例作为独立测试集，生成自动勾画结果。对比测试集上生成的自动勾画和“金标准”手动勾画的几何学差异［戴斯相似系数（DSC）、豪斯多夫距离（HD95）、体积差异（RAVD）、质心偏差（DC）］。将基于手动勾画制定的原始放疗计划映射至测试集生成自动勾画的结构上，评估临床靶区（CTV）和危及器官（OAR）上两种勾画方式效率（耗时）及关键剂量学参数差异，包括CTV参数：D₉₈、D₂、V₉₀、V₉₅、D_mean、HI；OAR参数：肠袋和直肠的V₃₀、V₄₀、V₅₀与D_mean；膀胱的V₅₀和D_mean，以及骨髓、双侧股骨头的D_mean。结果：AccuLearning系统自动勾画时间显著短于手动勾画（P<0.05）；自动勾画几何学参数结果显示，右侧股骨头DSC值最高，直肠DSC值最低，DSC均值均≥0.80；肠袋DC值最大，骨髓DC值最小。肠袋HD95值最大，骨髓HD95值最小。肠袋RAVD值最大，骨髓RAVD值最小；剂量学参数比较结果显示，自动勾画与手动勾画CTV的D₉₈、V₉₀、V₉₅、D_mean和HI差异具有统计学意义（P<0.05），在OAR方面，肠袋的V₄₀、V₅₀以及膀胱的V₅₀差异具有统计学意义（P<0.05）。结论：基于AccuLearning自动勾画系统显著提高宫颈癌勾画效率，几何学相似性高，对OAR具有较高的应用潜力，CTV仍需进一步修改以保持精度。

Abstract

Objective To evaluate the performance of the AccuLearning system for automated delineation in cervical cancer radiotherapy. Methods Planning CT images and manually delineated structures from 120 cervical cancer patients who received radiotherapy at the First Hospital of Lanzhou University between May 2022 and May 2025 were retrospectively included. Eighty cases were randomly selected as the training set to develop and train the automated delineation model, and the remaining 40 cases served as the independent test set to generate the automated delineation results. Geometric differences between automated delineation and gold-standard manual delineation from the test set were compared using the Dice similarity coefficient (DSC), 95% Hausdorff distance (HD95), relative absolute volume difference (RAVD), and deviation of centroid (DC). The original radiotherapy plan based on manual delineation was mapped to the automatically delineated contours from the test set. The efficiency (delineation time) and key dosimetric parameters of the two delineation methods for clinical target volume (CTV) and organs-at-risk (OAR) were evaluated. The assessed dosimetric parameters included CTV parameters (D ₉₈, D₂, V₉₀, V₉₅, D_mean, and HI) and OAR parameters (V₃₀, V₄₀, V₅₀ and D_mean for the intestinal pouch and rectum, V₅₀ and D_mean for the bladder, and the D_mean for the bone marrow and bilateral femoral head). Results The time required for automated delineation by the AccuLearning system was significantly shorter than that for manual delineation (P<0.05). Geometric evaluation of automated delineation results showed the maximum DSC value for the right femoral head and the minimum DSC value for the rectum, with all average DSC value ≥ 0.80. The intestinal pouch demonstrated the maximum CD, HD95, and RAVD, whereas bone marrow had the minimum CD, HD95, and RAVD. Dosimetric comparison revealed statistically significant differences in the D₉₈, V₉₀, V₉₅, D_mean and HI of CTV between automated delineation and manual delineation (P<0.05). For OAR, statistically significant differences were observed in V₄₀ and V₅₀ of the intestinal pouch and V₅₀ of the bladder (P<0.05). Conclusion Automated delineation using the AccuLearning system can significantly improve the delineation efficiency for cervical cancer and achieve high geometric similarity. Additionally, this system exhibits promising application potential for OAR, yet further optimization is required to preserve delineation accuracy for the CTV.

关键词

宫颈癌 / 放射治疗 / AccuLearning系统 / 自动勾画 / 手动勾画

Key words

cervical cancer / radiotherapy / AccuLearning system / automated delineation / manual delineation

参考文献

[1] 丁鑫, 王乾印, 马亮亮, 等 . 106例宫颈癌患者术后预后生存质量调查及预后COX回归分析［J］. 中国妇产科临床杂志, 2022, 23(3): 318-319.

[2] Ding X, Wang QY, Ma LL, et al. Investigation on postoperative prognostic quality of life and prognostic COX regression analysis in 106 cervical cancer patients［J］. Chinese Journal of Clinical Obstetrics and Gynecology, 2022, 23(3): 318-319.

[3] 胡静, 陈飞, 龚筱钦, 等 . 训练集病例数对基于深度学习宫颈癌临床靶区及危及器官自动勾画的影响［J］. 中国医疗设备, 2022, 37(9): 33-37.

[4] Hu J, Chen F, Gong XQ, et al. Effect of training volume on the automatic segmentation of clinical target volume and organs at risk in patients with cervical cancer based on deep learning［J］. China Medical Devices, 2022, 37(9): 33-37.

[5] 李霞, 刘娅, 王聪, 等 . 基于U—net卷积神经网络宫颈癌磁共振临床靶区和危及器官自动勾画的应用［J］. 中国医药导报, 2022, 19(24): 98-102.

[6] Li X, Liu Y, Wang C, et al. Application of automatic sketching of clinical target volume and organs at risk in cervical cancer based on U—net convolutional neural network［J］. China Medical Herald, 2022, 19(24): 98-102.

[7] 陈飞, 龚筱钦, 余云鹏, 等 . AccuLearning自动勾画临床靶区和危及器官用于宫颈癌术后放疗的可行性研究［J］. 实用医学杂志, 2024, 40(2): 153-157.

[8] Chen F, Gong XQ, Yu YP, et al. Feasibility of automatic segmentation of CTV and OARs in postoperative radiotherapy for cervical cancer using AccuLearning［J］. The Journal of Practical Medicine, 2024, 40(2): 153-157.

[9] 全科润, 柏朋刚, 陈文娟, 等 . 基于深度学习的宫颈癌放疗靶区及危及器官自动勾画研究［J］. 现代肿瘤医学, 2022, 30(20): 3759-3762.

[10] Quan KR, Bai PG, Chen WJ, et al. Automatic contouring of clinical target volume and organs at risk in radiotherapy for cervical cancer based on deep learning［J］. Journal of Modern Oncology, 2022, 30(20): 3759-3762.

[11] 曾志鹏, 左国平 . 基于nnUnetv2算法的宫颈癌放疗患者直肠壁自动勾画效果评价［J］. 现代肿瘤医学, 2024, 32(11): 2040-2045.

[12] Zeng ZP, Zuo GP . Effect evaluation of rectum wall automatic delineation based on nnUnetv2 algorithm in cervical cancer patients received radiotherapy［J］. Journal of Modern Oncology, 2024, 32(11): 2040-2045.

[13] Chen XM, Sun SL, Bai NRS, et al. A deep learning—based auto—segmentation system for organs—at—risk on whole—body computed tomography images for radiation therapy［J］. Radiother Oncol, 2021, 160: 175-184.

[14] Li YM, Rao S, Chen W, et al. Evaluating automatic segmentation for swallowing—related organs for head and neck cancer［J］. Technol Cancer Res Treat, 2022, 21: 15330338221105724.

[15] 程婷婷, 张子健, 杨馨, 等 . 基于集成学习的宫颈癌放射治疗危及器官的自动勾画［J］. 中南大学学报(医学版), 2022, 47(8): 1058-1064.

[16] Cheng TT, Zhang ZJ, Yang X, et al. Automatic delineation of organ at risk in cervical cancer radiotherapy based on ensemble learning［J］. Journal of Central South University(Medical Science), 2022, 47(8): 1058-1064.

[17] 黄新, 王新卓, 薛涛, 等 . 鼻咽癌放射治疗危及器官自动勾画的几何和剂量学分析［J］. 生物医学工程与临床, 2024, 28(1): 26-34.

[18] Huang X, Wang XZ, Xue T, et al. Geometric and dosimetric analysis of automatic segmentation of organs at risk for radiotherapy on nasopharyngeal carcinoma［J］. Biomedical Engineering and Clinical Medicine, 2024, 28(1): 26-34.

[19] 解治华, 路娜, 刘金锋, 等 . 儿童全骨髓全淋巴照射靶区和危及器官自动勾画［J］. 中国医学物理学杂志, 2024, 41(2): 163-168.

[20] Xie ZH, Lu N, Liu JF, et al. Auto—segmentation of target areas and organs—at—risk for total marrow and lymphoid irradiation in children［J］. Chinese Journal of Medical Physics, 2024, 41(2): 163-168.

[21] 陈飞, 胡静, 戴春华, 等 . 小样本训练模型在宫颈癌放疗中自动勾画可行性研究［J］. 中国医疗设备, 2021, 36(11): 27-31.

[22] Chen F, Hu J, Dai CH, et al. A feasibility study of automatic delineation using small sample data training algorithm model for the cervical cancer radiotherapy［J］. China Medical Devices, 2021, 36(11): 27-31.

[23] 左宇浩, 雷胜飞, 卢晓云, 等 . 基于卷积神经网络的腮腺浅叶自动勾画研究［J］. 医疗卫生装备, 2023, 44(5): 45-49.

[24] Zuo YH, Lei SF, Lu XY, et al. Automatic delineation of superficial parotid gland based on convolutional neural network［J］. Chinese Medical Equipment Journal, 2023, 44(5): 45-49.

[25] 齐英男, 陈雪梅, 陈佛平, 等 . 基于人工智能辅助勾画的宫颈癌放疗不同膀胱充盈度的剂量学研究［J］. 中国医学物理学杂志, 2025, 42(7): 847-852.

[26] Qi YN, Chen XM, Chen FP, et al. Dosimetric study on different bladder filling status in cervical cancer radiotherapy based on artificial intelligence—assisted segmentation［J］. Chinese Journal of Medical Physics, 2025, 42(7): 847-852.

[27] 陈美宁, 刘懿梅, 彭应林, 等 . 不同级别肿瘤中心医师对鼻咽癌调强放疗靶区和危及器官勾画差异比较［J］. 中国医学物理学杂志, 2024, 41(3): 265-272.

[28] Chen MN, Liu YM, Peng YL, et al. Comparison of interobserver variations in delineation of target volumes and organs—at—risk for intensity—modulated radiotherapy of nasopharyngeal carcinoma among physicians from different levels of cancer centers［J］. Chinese Journal of Medical Physics, 2024, 41(3): 265-272.

[29] 徐广庆, 蔡文涛, 徐丽丽, 等 . 基于人工智能技术的左侧乳腺癌放疗患者危及器官自动勾画的几何和剂量学精度研究［J］. 哈尔滨医科大学学报, 2024, 58(2): 143-148.

[30] Xu GQ, Cai WT, Xu LL, et al. Geometric and dosimetric accuracy research of auto—segmentation organs at risk in patients with left breast cancer radiotherapy based on artificial intelligence technology［J］. Journal of Harbin Medical University, 2024, 58(2): 143-148.

[31] 鲜利勋, 李光俊, 肖青, 等 . 基于食管癌放疗的剂量学参数研究几何指标评价自动勾画轮廓准确度的可行性［J］. 中国医学物理学杂志, 2021, 38(2): 148-152.

[32] Xian LX, Li GJ, Xiao Q, et al. Feasibility study of evaluating autosegmentation accuracy with geometric indices based on dosimetric parameters of esophageal cancer radiotherapy［J］. Chinese Journal of Medical Physics, 2021, 38(2): 148-152.

[33] 李陆军, 游雁, 谢金莲, 等 . AI勾画危及器官对鼻咽癌放疗计划剂量优化的影响［J］. 现代肿瘤医学, 2023, 31(15): 2899-2903.

[34] Li LJ, You Y, Xie JL, et al. Application of artificial intelligence to auto—segmentation organ at risk in radiotherapy for nasopharyngeal carcinoma［J］. Journal of Modern Oncology, 2023, 31(15): 2899-2903.

[35] 陈旭明, 刘勇 . 基于深度学习的正常组织自动勾画在计划设计中的剂量准确度评估［J］. 中国医疗设备, 2021, 36(10): 169-172.

[36] Chen XM, Liu Y . Evaluation of the dose accuracy in application of deep Learning—Based automatic delineation of normal tissues in treatment planning［J］. China Medical Devices, 2021, 36(10): 169-172.

[37] 田磊, 薛晓英, 王艳强, 等 . 宫颈癌术后保护骨髓调强放疗的剂量学优势［J］. 河北医科大学学报, 2021, 42(10): 1210-1214.

[38] Tian L, Xue XY, Wang YQ, et al. Dosimetric advantages of bonemarrow—sparing intensity modulated radiotherapy for cervical cancer after hysterectomy［J］. Journal of Hebei Medical University, 2021, 42(10): 1210-1214.

基金资助

甘肃省科技计划基础研究计划(25JRRA557)