两种软件Smart Segmentation与MIM Atlas自动勾画鼻咽癌危及器官的准确性研究

目的 比较两种自动勾画软件（Smart Segmentation与MIM Atlas）勾画鼻咽癌危及器官（OAR）的准确性。方法 回顾性选取2015-2016年浙江大学医学院附属第一医院收治的鼻咽癌患者共55例,在CT图像上手动勾画OAR,以简单随机抽样方式取其中30例在Smart Segmentation与MIM Atlas中创建各自的病例库,剩余25例作为测试病例在两个软件中运行得到两组自动勾画结果。以手动勾画为金标准,计算两组自动勾画结果的戴斯相似性系数（DSC）、豪斯多夫距离（HD）、绝对体积差（△V）,通过比较以上3个参数来评估两种软件勾画鼻咽癌危及器官的准确性。结果 Smart Segmentation与MIM Atlas勾画所有器官的总体DSC分别为（0.79±0.13）和（0.62±0.24）（t=14.06,P<0.05）;总体HD分别为（5.50±3.84）和（8.38±4.88）mm（t=-11.40,P<0.05）;总体△V为（1.52±2.46）、（2.38±3.57）cm³（t=-4.70,P<0.05）。MIM Atlas勾画的11个器官（脑干、视交叉、左右眼晶状体、左右视神经、左右眼球、左右侧腮腺、脊髓）的DSC均值大于Smart Segmentation的结果（t=5.27、4.41、6.34、5.70、10.62、7.45、3.96、4.26、6.25、5.42、7.23,P<0.05）。MIM Atlas勾画的10个器官（脑干、视交叉、左右眼晶状体、左右视神经、左右眼球、左侧腮腺、脊髓）的HD均值小于Smart Segmentation（t=-4.51、-4.49、-3.92、-3.45、-5.36、-5.56、-3.89、-3.90、-3.60、-3.68,P<0.05）。MIM Atlas勾画的6个器官（脑干、视交叉、左眼晶状体、左右视神经、右眼球）的△V均值小于Smart Segmentation（t=-2.83、-3.39、-2.56、-2.27、-2.43、-2.51,P<0.05）。结论 对于体积较大的器官,两种软件都有较好的勾画结果。器官的体积越小、边界越模糊,则勾画结果越差。MIM Atlas的勾画结果总体上优于Smart Segmentation。     

Automatic segmentation of thoracic CT images using three deep learning models

PurposeDeep learning (DL) techniques are widely used in medical imaging and in particular for segmentation. Indeed, manual segmentation of organs at risk (OARs) is time-consuming and suffers from inter- and intra-observer segmentation variability. Image segmentation using DL has given very promising results. In this work, we present and compare the results of segmentation of OARs and a clinical target volume (CTV) in thoracic CT images using three DL models.Materials and methodsWe used CT images of 52 patients with breast cancer from a public dataset. Automatic segmentation of the lungs, the heart and a CTV was performed using three models based on the U-Net architecture. Three metrics were used to quantify and compare the segmentation results obtained with these models: the Dice similarity coefficient (DSC), the Jaccard coefficient (J) and the Hausdorff distance (HD).ResultsThe obtained values of DSC, J and HD were presented for each segmented organ and for the three models. Examples of automatic segmentation were presented and compared to the corresponding ground truth delineations. Our values were also compared to recent results obtained by other authors.ConclusionThe performance of three DL models was evaluated for the delineation of the lungs, the heart and a CTV. This study showed clearly that these 2D models based on the U-Net architecture can be used to delineate organs in CT images with a good performance compared to other models. Generally, the three models present similar performances. Using a dataset with more CT images, the three models should give better results.     

基于磁共振图像手工勾画宫颈癌危及器官稳定性分析

目的 以宫颈癌MR图像为基础,分析不同勾画者、同一勾画者不同时间手工勾画危及器官（OARs）的稳定性,为放疗计划设计的序列选择进行初步探索。方法 回顾性分析中山大学肿瘤防治中心放疗科2016—2018年经MR-sim扫描的30例宫颈癌患者,选取共有的T1WI、T1dixonc和T2WI 3序列MR图像导入Monaco计划系统,由两位临床放疗医师独立在每个患者3个序列上分别勾画膀胱、直肠、肛管、左/右股骨头。其中一位医师在完成首次勾画工作后的一个月再次完成T1WI序列及各OARs的勾画。统计分析各OARs的相似性系数（DSC）、豪斯多夫距离（HD）和位置差异（Δx、Δy、Δz）。结果 不同勾画者在T1WI、T1dixonc、T2WI序列和同一个勾画者在T1WI序列上勾画5个OARs的HD值均<2 mm;位置差异均<5 mm。不同勾画者和同一个勾画者不同时间勾画的DSC、HD及位置差异与OARs体积呈正相关性（R=0.178~0.582,P<0.05）。因肛管体积较小（7.385±1.555）cm³,DSC值均<0.7表现稍差外,其余OARs平均DSC值均>0.82。通过两两比较3个序列上不同勾画者勾画OARs的DSC、HD发现,T1WI序列直肠、左/右股骨头的DSC值、膀胱、左/右股骨头、直肠的HD值以及肛管、右股骨头Δz轴差异均优于T1dixonc,差异有统计学意义（t=-3.116~3.604,P<0.05）;T1WI序列直肠DSC值和肛管HD值较T2WI序列好（t=2.934、3.677,P<0.05）;T1dixonc序列直肠DSC、肛管HD差异稍优于T2WI（t=6.806、2.130,P<0.05）;T2WI序列勾画骨组织（左/右股骨头）稳定性优于T1WI、T1dixonc,且差异均具有统计学意义（t=-6.580~6.542,P<0.05）。结论 基于MR图像的不同勾画者和同一勾画者不同时间勾画膀胱、直肠、股骨头稳定性较好,肛管次之。且T1WI序列OARs勾画稳定性较优于T1dixonc、T2WI。     

Prospective randomized double-blind study of atlas-based organ-at-risk autosegmentation-assisted radiation planning in head and neck cancer

Background and purpose

Target volumes and organs-at-risk (OARs) for radiotherapy (RT) planning are manually defined, which is a tedious and inaccurate process. We sought to assess the feasibility, time reduction, and acceptability of an atlas-based autosegmentation (AS) compared to manual segmentation (MS) of OARs.

Materials and methods

A commercial platform generated 16 OARs. Resident physicians were randomly assigned to modify AS OAR (AS + R) or to draw MS OAR followed by attending physician correction. Dice similarity coefficient (DSC) was used to measure overlap between groups compared with attending approved OARs (DSC = 1 means perfect overlap). 40 cases were segmented.

Results

Mean ± SD segmentation time in the AS + R group was 19.7 ± 8.0 min, compared to 28.5 ± 8.0 min in the MS cohort, amounting to a 30.9% time reduction (Wilcoxon p < 0.01). For each OAR, AS DSC was statistically different from both AS + R and MS ROIs (all Steel–Dwass p < 0.01) except the spinal cord and the mandible, suggesting oversight of AS/MS processes is required; AS + R and MS DSCs were non-different. AS compared to attending approved OAR DSCs varied considerably, with a chiasm mean ± SD DSC of 0.37 ± 0.32 and brainstem of 0.97 ± 0.03.

Conclusions

Autosegmentation provides a time savings in head and neck regions of interest generation. However, attending physician approval remains vital.