摘要
目的:评估基于U形全卷积神经网络(U-Net)对CT图像上Couinaud法肝段的自动分割及体积测量的准确性,探讨其用于半肝切除术剩余肝脏体积百分比(FLR%)评估的可行性。方法:回顾性收集医学中心A的腹部CT增强扫描数据(共170例)用于肝段分割模型的建立,先分割肝脏轮廓,再训练自动分割肝段的模型,最终获得各肝段体积。将放射科医生标注的肝段数据作为金标准。采用医学中心B的CT数据(50例)作为外部验证集,以平均Dice相似性系数(DSC)评价模型效能,分析比较模型与医师在肝段分割、体积测量、FLR%评估上的差异。结果:医学中心A所有数据随机分为训练集(132例)、调优集(19例)、测试集(19例)。外部验证集平均DSC值为(0.92±0.00),肝段平均体积最小为Ⅰ段[(37.59±1.26) mL],最大为Ⅷ段[(241.76±6.07) mL]。模型与手工标注FLR%评估结果一致性高(95%一致范围为0.9768~0.9906),在手术可行性预测上差异无统计学意义(P=0.25)。结论:基于U-Net的Couinaud’s肝段自动分割、体积测量并评估半肝切除术FLR%具有可行性。
Objective:To assess the accuracy of the automated segmentation and volume calculation of Couinaud’s liver segment on CT images by using the U-Net segmentation algorithm, and to assess the feasibility of this algorithm in the assessment of future liver remnant percentage(FLR%) prior to hemi-hepatectomy.Methods:Abdominal-enhanced CT images of 170 patients in medical center A were retrospectively collected and used for the development of the automated segmentation model.First, the liver contour was segmented automatically, then the model for automatic sub-segmentation of Couinaud’s segment was trained, and the volume of each segment was calculated.Couinaud’s segmentations which were annotated by radiologists were regarded as the ground truth.CT images of 50 patients in medical center B were used for external testing.The Dice similarity coefficient was used to assess the performance of the model.And the segmentation results of Couinaud’s segment, volume calculation, and FLR% evaluation for hemi-hepatectomy between the model and doctor were compared.Results:All data in medical center A were randomly divided into the dataset of the train(n=132),validate(n=19),and test(n=19).The mean Dice similarity coefficient was(0.92±0.00) in the external testing group.For the average volume of Couinaud’s segment, the smallest was Segment Ⅰ(37.49±1.26mL) and the largest was Segment Ⅷ(241.76±6.07mL).FLR % assessments of automated and manual segmentation agreed closely(95% limits of agreement: 0.9768 to 0.9906).In the qualitative analysis, no statistical difference was found between the model and the doctor on the permission of hemi-hepatectomy(P=0.25).Conclusion:Fully automated segmentation and volumetric analysis of Couinaud’s liver segment, and FLR% evaluation prior to hemi-hepatectomy is feasible by using the U-Net segmentation algorithm.
作者
谢婷婷
刘想
林子楹
张晓东
张耀峰
张大斗
成官迅
王霄英
XIE Ting-ting;LIU Xiang;LIN Zi-ying(Department of Radiology,Peking University First Hospital,Beijing 100034,China)
出处
《放射学实践》
CSCD
北大核心
2023年第1期47-51,共5页
Radiologic Practice
基金
北京大学深圳医院科研基金资助课题(JCYJ2020007)。
