目的 评估人工智能技术在左侧乳腺癌患者危及器官自动勾画中的几何和剂量学精度,为自动勾画技术的临床应用提供参考.方法 选取2021 年4 月～2022 年 6 月于哈尔滨医科大学附属第一医院行左侧乳腺癌放射治疗患者13 例的资料作为研究对象.应用人工智能勾画软件对危及器官(organ at risk,OAR)进行手动和自动勾画,其中OAR包括:左右肺、心脏、气管、食道和脊髓,比较两组勾画间的Dice相似系数(dice similarity coefficient,DSC)、Jaccard系数(Jaccard coefficient,JC)、Hausdorf距离(Haus-dorff distance,HD)、质心距离(center of mass deviation,CMD)、包容性系数(inclusive index,IncI)、敏感性指数(sensitivity index,SI)和勾画时间.在剂量学精度方面,提取两组勾画方式的左肺V5、V10、V20、V30、Dmean、右肺Dmean、心脏V20、V25、V30、V40、Dmean、脊髓Dmax、食道Dmax和气管Dmax剂量学数据,比较剂量学数据间的统计学差异情况.结果 几何精度方面,所有OAR的DSC≥0.83,JC≥0.70,HD≤17.06 mm,CMD≤3.03 mm,IncI≥0.74,SI≥0.89.剂量学数据方面,手动勾画和自动勾画的主要剂量学指标,左肺V20分别为(21.39±5.19)%和(21.24±4.93)%,心脏V25分别为(9.57±4.36)%和(9.28±4.15)%,脊髓Dmax分别为(19.98±12.59)Gy和(19.92±12.65)Gy,均满足临床对乳腺癌放疗OAR的保护要求,且两组勾画间的剂量学数据差异无统计学意义(P＞0.05).自动勾画和手动勾画的时间分别为(65.23±7.01)s vs.(1 808.92±104.03)s(P＜0.05).结论 自动勾画软件能以较高的几何和剂量学精度完成左侧乳腺癌放疗患者的OAR勾画工作,同时节省了大量勾画时间,提高了勾画效率.
Geometric and dosimetric accuracy research of auto-segmentation organs at risk in patients with left breast cancer radiotherapy based on artificial intelli-gence technology
Objective To evaluate the geometric and dosimetric accuracy of artificial intelli-gence technology in the auto-segmentation of left breast cancer organs at risk(OAR),and to provide reference for the clinical application of auto-segmentation technology.Methods The data of 13 patients with left breast cancer treated with radiotherapy in the First Affiliated Hospi-tal of Harbin Medical University from April 2021 to June 2022 were selected.The OAR were contoured manually and automatically by artificial intelligence delineation software respectively.The OAR included left and right lungs,heart,trachea,esophagus and spinal cord.The dice similarity coefficient(DSC)and Jaccard coefficient(JC),Hausdorf distance(HD),center of mass distance(CMD),inclusive coefficient(IncI),sensitivity index(SI),and delineation time between the two groups were compared.In terms of dosimetric accuracy,the dosimetric data of left lung V5,V10,V20,V30,Dmean,right lung Dmean,heart V20,V25,V30,V40,Dmean,spinal cord Dmax,esophagus Dmax and trachea Dmax were extracted from the two groups of delineated methods to compare the statistical differences between the dosimetry data.Results In terms of geometric accuracy,DSC≥0.83,JC≥0.70,HD≤17.06 mm,CMD≤3.03 mm,IncI≥0.74,SI≥0.89 of all OARs.In terms of dosimetric data,the main dosimetric indicators of manual delineation and auto-segmentation,the left lung V20 were(21.39±5.19)%and(21.24±4.93)%,heart V25 were(9.57±4.36)%and(9.28±4.15)%,and spinal cord Dmax were(19.98±12.59)Gy and(19.92±12.65)Gy,respectively.They both meet the clinical OAR protection requirements for breast cancer radiotherapy,and there was no signifi-cant difference in the dosimetry data between the two groups(P＞0.05).The times of auto-matic and manual delineation were(65.23±7.01)s VS(1 808.92±104.03)s(P＜0.05),respectively.Conclusion The auto-segmentation software can complete the delineation of OAR in patients with left breast cancer radiotherapy with high geometric and dosimetric accura-cy,while saving a lot of delineation time,and improving delineation efficiency.
artificial intelligenceleft breast cancerorgans at riskauto-segmentationgeo-metric and dosimetric accuracy
NETL
NSTL
万方数据
