摘要
目的:评估基于患者图谱库的自动轮廓勾画软件(ABAS)在鼻咽癌调强计划中勾画危及器官的准确性及临床应用效果。方法:选取已使用调强治疗的10例鼻咽癌患者,将每例患者的定位CT图像及医生手动勾画的调强计划中关注的13种(20个)危及器官传人ABAS软件作为数据库模板。本研究选取22例新患者应用ABAS进行自动勾画.其结果传回计划系统并由各主管医生在此基础上进行手动修改。对ABAS自动勾画结果与医生在此基础上修改后的结果进行比较,并对使用时间进行统计。结果:ABAS自动勾画与医生修改勾画的形状相似性系数(DSC)平均值超过0.9的为下颌骨(0.97±0.02),颞叶(左0.97±0.03,右0.97±0.03),脑干(0.94±0.04),喉(0.94±0.06),脊髓(0.93±0.05),腮腺(左0.91±0.07,右0.91±0.08),眼球(左0.90±0.13,右0.96±0.13);DSC系数低于0.9的为颞颌关节(左0.89±0.10,右0.87±0.14),内耳(左O.86±0.14,右0.83±0.16),垂体(0.83±0.10),视交叉(0.70±0.14),晶体(左0.69±0.11,右0.70±0.14),视神经(左0.65±0.25。右0.64±0.14)。使用ABAS自动产生所有危及器官的时间约10分钟/每患者;医生在自动勾画基础上修改时间为6~25min/每患者,平均约15min/每患者。以前完全手动勾画以上危及器官估计需90min到120min,使用ABAS自动勾画则明显的缩短了时间。结论:基于患者图谱库的ABAS自动勾画软件,对目前鼻咽癌患者调强计划关注的大部分危及器官能够达到满意的自动勾画结果,临床应用表明能明显的节约医生勾画时间。
Objective: To evaluateauto-contouring accuracy and clinical efficiency using atlas-based auto-segmentation (ABAS) of CT images for nasopharyngeal (NPC)patient in intensity-modulated radiotherapy (IMRT). Methods: The images from 10 NPC patients treated with IMRT were selected as atlases input for ABAS, which including 20 organs-at-risk (OARs) contoured by doctors. For 22 clinical new patients, theauto contours generated by ABAS were compared with that edited auto contours by doctors. Dice similarity coeff±cients (DSC) were calculated between the auto contours and edited autocontours. The editing times were recorded. In addition, doctors conducted a subjective evaluation about the editing degree for each OAR. Results: The mean DSC greater than 0.9 were mandible (0.97+0.02), temporal lobe (Left (L) 0.97±0.03, Right(R) 0.97±0.03), brainstem (0.94±0.04), larynx (0.94±0.06), spinalcord (0.93±0.05),parotid (L 0.91 ±0.07, R 0.91±0.08), eye (L 0.90±0.13, R 0.96±0.13). DSC less than 0.9 were temporomandibular joint (L 0.89±0.10, R 0.87±0.14), inner ear (L 0.86±0.14, R 0.83 ±0.16), pituitary (0.83±0.10), optic chiasm (0.70±0.14), Lens (L 0.69±0.11, R 0.70±0.14), optic nerve (L 0.65±0.25, R 0.64±0.14). The time to auto-segment all OARs using ABAS was about 10 minutes per patient. Theediting time ranged from 6 to 25 minutes per patient, and the mean editing time was 15 minutes. Comparing the fully manual contour time of 90 to 120 minutes,ABAS auto-contour significantly shorten the time.Conclusions: Automatic segmentation generates contours of sufficient accuracy for most OARs in clinical NPC IMRT plan. The clinic.al application shows thatsubstantial time saving was achieved using editing, instead of manual contouring (15 vs.〉90min). In combination with other methods to improve automatic segmentation accuracy and broad applicability is under further study.
出处
《中国医学物理学杂志》
CSCD
2013年第2期3997-4000,4035,共5页
Chinese Journal of Medical Physics