论文部分内容阅读
将卷积神经网络(CNN)用于肺部肿瘤正电子发射计算机断层扫描(PET)/电子计算机断层扫描(CT)计算机辅助诊断,不仅可以提供精确的定量分析以弥补人眼惰性及对灰阶不敏感的缺陷,也能辅助医生准确诊疗。本文首先采用参数迁移的方法初始构建三个CNN(CT-CNN、PET-CNN、PET/CT-CNN)分别用于肺部肿瘤CT、PET、PET/CT的识别;然后以CT-CNN为例探讨迭代次数、批次大小和输入图像大小对识别率和训练时间的影响,从而选择合适的模型参数训练单一CNN;最后集成三个单一CNN,采用“相对多数投票法”完成肺部肿瘤PET/CT计算机辅助诊断,进而对比集成CNN与单个CNN的性能。实验结果表明集成CNN模型比单一CNN模型对于肺部肿瘤计算机辅助诊断的性能更优。
Using the Convolutional Neural Network (CNN) for Pulmonary Tumor Positron Emission Tomography (PET) / computed tomography (CT) computer-aided diagnosis can not only provide accurate quantitative analysis to compensate for the inertia of the human eye but also the effect on grayscale Insensitive defects, but also to assist doctors accurate diagnosis and treatment. In this paper, three CNNs (CT-CNN, PET-CNN, PET / CT-CNN) were initially constructed for the identification of CT, PET and PET / CT of lung tumors using parameter migration method. The effect of iteration number, batch size and input image size on recognition rate and training time is discussed to select the appropriate model parameters to train a single CNN. Finally, three single CNNs are integrated and the “majority vote method” PET / CT computer-aided diagnosis, and then compare the performance of integrated CNN with a single CNN. The experimental results show that the integrated CNN model is better than the single CNN model for computer-aided diagnosis of lung tumors.