摘要
目前,癌症仍然是人类身体健康的一个巨大威胁。癌症的诊断尤其是早期诊断,对于癌症的治愈来说至关重要。以正电子发射型计算机断层显像/计算机断层扫描显像(Positron Emission Computed Tomography/Computed Tomography,PET/CT)为代表的医疗影像学技术是癌症早期诊断的重要方法。随着计算机技术的发展,尤其是深度学习技术的出现,模型可以自动从PET/CT图像中提取有效的特征,使得对癌症进行早期诊断成为可能。然而,基于以卷积神经网络为代表的深度学习技术搭建的模型依赖大量数据的训练,同时PET/CT检查费用过于高昂且给病人带来的生理不适过强,目前PET/CT设备并未得到大范围推广,而可用的PET/CT数据数量不大。因此,可以使用生成对抗网络(Generative Adversarial Network,GAN)对已有的PET/CT数据进行数据扩充,即依据已有的真实PET/CT图像生成大量仿真PET/CT图像,再将扩充后的数据集用于癌症早期诊断模型的训练。通过搭建一个基于CycleGAN的模型证明:使用CycleGAN模型生成的PET/CT图像基本符合使用要求,可有效扩充数据集。
So far,cancer is still a great threat to human health.The diagnosis of cancer,especially the early diagnosis,is very important for the cure of cancer.Positron Emission Computed Tomography/Computed Tomography(PET/CT)is an important method for early diagnosis of cancer.With the development of computer technology,especially the emergence of deep learning technology,it is possible for models to automatically extract effective features from PET/CT images and predict cancer early.However,the model based on deep learning technology represented by convolutional neural network relies on a large number of data training.Due to the high cost of PET/CT examination and the strong physiological discomfort to patients,the current PET/CT equipment has not been widely promoted,and the available PET/CT data set is not large.Therefore,Generative Adversarial Networks(GAN)can be used to expand the existing PET/CT data,that is,a large number of simulated PET/CT images can be generated based on the existing real PET/CT images,and then the expanded data can be used to train the cancer diagnosis model.By building a model based on CycleGAN,it is proved that the PET/CT images generated by CycleGAN model basically meet the use requirements,and the method of generating simulation PET/CT images to expand data sets based on CycleGAN is feasible and effective.
作者
李柏桥
彭显
LI Boqiao;PENG xian(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming,650050,China)
出处
《电视技术》
2021年第3期117-121,共5页
Video Engineering
