Technical Note: A deep learning-based autosegmentation of rectal tumors in MR images
Technical Note: A deep learning-based autosegmentation of rectal tumors in MR images
复制标题
基于深度学习的 MR 图像中直肠肿瘤自动分割
DOI：
10.1002/mp.12918
复制
发表时间：
2018-06-01
期刊：
影响因子：
3.8
通讯作者：
Hu, Weigang
中科院分区：
文献类型：
--
作者：
Wang, Jiazhou;Lu, Jiayu;Hu, Weigang
目的：在直肠癌放射治疗中，手动勾画大体肿瘤体积（GTV）是一个关键而耗时的过程。本研究旨在开发一种简单的基于深度学习的自动分割算法来分割T2加权MR图像上的直肠肿瘤。材料和方法：本研究纳入了93例接受新辅助放化疗后手术治疗的局部晚期（cT 3 -4和/或cN 1 -2）直肠癌患者的MRI扫描（3 T，T2加权）。建立了一个二维U网类神经网络作为训练模型。该模型分为两个阶段进行训练，以提高效率。这些阶段是肿瘤识别和肿瘤分割。分割后的轮廓进行了一个开放（腐蚀和膨胀）的过程。数据被随机分为训练（90%）和验证（10%）数据集，用于10个文件夹的交叉验证。此外，对20例患者进行了双轮廓成形以进行性能评价。计算Hausdorff距离（HD）、平均表面距离（ASD）、Dice指数（DSC）和Jaccard指数（JSC）4个指标来评价自动分割和手动分割的相似性。DSC、JSC、HD和ASD对于验证数据集，（平均值+/- SD）分别为0.74 +/- 0.14、0.60 +/- 0.16、20.44 +/- 13.35和3.25 +/- 1.69 mm;两名人类放射肿瘤学家之间的这些指数分别为0.71 ± 0.13、0.57 ± 0.15、14.91 ± 7.62和2.67 ± 1.46 mm。考虑到DSC（P = 0.42）、JSC（P = 0.35）、HD（P = 0.079）和ASD（P = 0.16），在自动分割和手动分割之间未观察到显著差异。结论：本研究表明，简单的深度学习神经网络可以基于MRI T2图像对直肠癌进行分割，结果与人类相当。(c)2018年美国医学物理学家协会
Purpose: Manual contouring of gross tumor volumes (GTV) is a crucial and time-consuming process in rectum cancer radiotherapy. This study aims to develop a simple deep learning-based autosegmentation algorithm to segment rectal tumors on T2-weighted MR images.Material and methods: MRI scans (3T, T2-weighted) of 93 patients with locally advanced (cT3-4 and/or cN1-2) rectal cancer treated with neoadjuvant chemoradiotherapy followed by surgery were enrolled in this study. A 2D U-net similar network was established as a training model. The model was trained in two phases to increase efficiency. These phases were tumor recognition and tumor segmentation. An opening (erosion and dilation) process was implemented to smooth contours after segmentation. Data were randomly separated into training (90%) and validation (10%) datasets for a 10-folder cross-validation. Additionally, 20 patients were double contoured for performance evaluation. Four indices were calculated to evaluate the similarity of automated and manual segmentation, including Hausdorff distance (HD), average surface distance (ASD), Dice index (DSC), and Jaccard index (JSC).Results: The DSC, JSC, HD, and ASD (mean +/- SD) were 0.74 +/- 0.14, 0.60 +/- 0.16, 20.44 +/- 13.35, and 3.25 +/- 1.69 mm for validation dataset; and these indices were 0.71 +/- 0.13, 0.57 +/- 0.15, 14.91 +/- 7.62, and 2.67 +/- 1.46 mm between two human radiation oncologists, respectively. No significant difference has been observed between automated segmentation and manual segmentation considering DSC (P = 0.42), JSC (P = 0.35), HD (P = 0.079), and ASD (P = 0.16). However, significant difference was found for HD (P = 0.0027) without opening process.Conclusion: This study showed that a simple deep learning neural network can perform segmentation for rectum cancer based on MRI T2 images with results comparable to a human. (c) 2018 American Association of Physicists in Medicine