基于密集全连接卷积网络的宫颈癌患者CTV自动预勾画

Automated Pre-delineation of CTV in Patients with Cervical Cancer Using Dense V-Net

摘要: 使用在小样本中特征学习效果理想的密集全连接卷积网络基于CT图像对宫颈癌患者CTV进行自动预勾画，并评估效果。研究随机选取勾画范围相近的IB期和IIA期宫颈癌术后患者CT数据，从勾画相似度、轮廓偏移程度和勾画体积差异程度三个方面评估预勾画精度。经验证，8项最具代表性的参数结果优于单一网络勾画结果，并优于国际报道。Dense V-Net可较为准确地实现宫颈癌患者CTV预勾画，经医生审查简单修改，可用于临床。

Abstract: We use a dense and fully connected convolutional network with good feature learning in small samples, to automatically pre-deline CTV of cervical cancer patients based on CT images and evaluate the effect. The CT data of stage IB and IIA postoperative cervical cancer with similar delineation scope were selected to be used to evaluate the pre-sketching accuracy from three aspects:sketching similarity, sketching offset and sketching volume difference. It has been proved that the 8 most representative parameters are superior to those with single network and reported internationally before. Dense V-Net can accurately predict CTV pre-delineation of cervical cancer patients, which can be used clinically after simple modification by doctors.