摘要
目的探究在鼻咽癌自适应放疗中,应用基于配准对抗神经网络的RegGAN模型将锥形束CT(CBCT)图像转换为伪CT图像的可行性。方法回顾60例鼻咽癌患者第1次用于图像引导放疗的CBCT图像和计划CT图像,其中47例作为训练集,13例作为测试集。建立基于配准对抗神经网络的RegGAN模型,将CBCT图像转换为伪CT图像。选用Pix2Pix模型和Cycle-consistency模型作为参考模型,在相同训练集上进行模型训练后使用测试集进行图像转换。比较利用3种模型转换的伪CT图像的平均绝对误差(MAE)、峰值信噪比(PSNR)和结构相似性(SSIM)。结果在测试集上应用RegGAN模型转换的伪CT图像具有更多的纹理信息。与Pix2Pix模型和Cycle-consistency模型相比,应用RegGAN模型转换的伪CT图像的MAE最小,PSNR和SSIM最大(均P<0.05)。结论采用RegGAN模型从CBCT图像转换而来的伪CT图像具有较高的图像质量,可以作为鼻咽癌自适应放疗的参考图像。
Objective To explore the feasibility of converting cone beam CT(CBCT)image to simulated CT image by the registration adversarial neural network-based RegGAN model in adaptive radiotherapy for nasopharyngeal carcinoma.Methods Sixty patients with nasopharyngeal carcinoma were enrolled to review their CBCT images and plan CT images which were used for image-guided radiotherapy for the first time,including 47 cases serving as the training set and 13 cases as the test set.The registration adversarial neural network-based RegGAN model was established to convert CBCT images to simulated CT images.Using the Pix2Pix model and the Cycle-consistency model as the reference models,image converting was performed on the test set following model training in the same training set.Mean absolute error(MAE),peak signal-to-noise ratio(PSNR),and structural similarity(SSIM)were compared among the simulated CT images converted by the three different models.Results On the test set,the simulated CT images converted by the RegGAN model had more texture information.Compared with the Pix2Pix model and the Cycle-consistency model,the simulated CT images converted by the RegGAN model exhibited the minimum MAE and the maximum PSNR and SSIM(all P<0.05).Conclusion The simulated CT image converted from CBCT image using the RegGAN model shows higher quality of image,and can serve as a reference image for adaptive radiotherapy in nasopharyngeal carcinoma.
作者
刘培
朱超华
孔令轲
廖超龙
陆合明
LIU Pei;ZHU Chao-hua;KONG Ling-ke;LIAO Chao-long;LU He-ming(Youjiang Medical University for Nationalities,Baise 533000,China;Department of Radiotherapy,the People′s Hospital of Guangxi Zhuang Autonomous Region,Nanning 530021,China;Manteia Technologies Co.,Ltd.,Xiamen 361008,China)
出处
《广西医学》
CAS
2021年第20期2397-2400,2405,共5页
Guangxi Medical Journal
基金
广西重点研发计划(桂科AB17195005)。
关键词
配准对抗神经网络
RegGAN模型
锥形束CT
伪CT
图像转换
图像质量
深度学习
鼻咽癌
自适应放疗
Registration adversarial neural network
RegGAN model
Cone beam CT
Simulated CT
Image conversion
Image quality
Deep learning
Nasopharyngeal carcinoma
Adaptive radiotherapy