基于增强CT的手动勾画与基于PET-CT的自动勾画方法在鼻咽癌放疗中的对比研究

目的:探究基于增强CT的手动勾画与基于PET-CT的自动勾画方法在鼻咽癌放疗中的优劣性及应用价值。方法:采集34例鼻咽癌患者的PET-CT数据，比较医生分别在增强CT和PET-CT图像上勾画靶区的差异性，以CSCO指南在PET-CT双模态融合图像上手动勾画的靶区作为金标准，应用Matlab软件计算单一标准摄取值（SUV）阈值分割法和MIM软件自带的PET-EDGE分割法两者的重叠率Dice和豪斯多夫距离HD值，并采用t检验进行统计学分析。结果:基于增强CT图像的手动勾画法、基于PET-CT图像的单一SUV阈值法以及PET-EDGE自动勾画法，3者靶区分割结果的Dice值分别为0.706、0.736、0.806，HD值分别为12.601、11.631、6.143 mm。以PET-CT双模态图像作为参考的靶区手动勾画法与基于增强CT的手动勾画存在差异（Dice:P=8.33×10-7；HD:P=1.065×10-18）；MIM软件自带的PET-EDGE靶区分割法Dice和HD值高于单一SUV阈值法，差异具有统计学意义（Dice:P=0.019 8；HD:P=0.013 7）。结论:在鼻咽癌肿瘤靶区的分割性能上，基于PET-CT图像的自动勾画优于基于增强CT图像的手动勾画，MIM软件自带的PET-EDGE分割法优于单一SUV阈值分割法。

Comparison of manual segmentation in enhanced CT image and automatic segmentation in PET-CT image for nasopharyngeal carcinoma radiotherapy

Abstract: Objective To explore the pros and cons of manual segmentation in enhanced CT image and automatic segmentation in PET-CT image for radiotherapy for nasopharyngeal carcinoma, and to discuss their application value. Methods The PET-CT data of 34 patients with nasopharyngeal carcinoma was collected. The performance of target segmentation in enhanced CT and PET-CT images were compared. The tumor area manually drawn by the doctor in PET-CT image according to the CSCO guidelines was taken as the ground truth, and the Dice similarity coefficient and Hausdorff distance (HD) of the automatic segmentation by SUV threshold or PET-EDGE of MIM software were calculated by Matlab software. Moreover, t-test was used for statistical analysis. Results The Dice of manual segmentation in enhanced CT images, automatic segmentation by SUV threshold in PET-CT image, and automatic segmentation by PET-EDGE in PET-CT image were 0.706, 0.736, and 0.806, respectively, and the HD was 12.601, 11.631, and 6.143 mm, respectively. There were differences between manual segmentation in PET-CT image and manual segmentation in enhanced CT image (Dice: P=8.33×10-7 HD: P=1.065×10-18). The Dice and HD of automatic segmentation by PET-EDGE of MIM software were higher as compared with the automatic segmentation by SUV threshold, and the differences were statistically significant (Dice: P=0.0198 HD: P=0.0137). Conclusion The automatic segmentation in PET-CT image is advantageous over manual segmentation in enhanced CT image in target segmentation for nasopharyngeal carcinoma, and PET-EDGE of MIM software is superior to SUV threshold.