基于U-Net网络的肺部CT图像中的肺结节轮廓识别
首发时间:2019-11-29
摘要:在肺部CT图像切片上,肺结节很小并且有些肺结节与周围良性组织相连。由于肺结节的灰度值与周围良性组织的灰度值相差很小,这使得肺结节轮廓识别有很大难度。为了完成肺结节轮廓识别任务,设计了基于U-Net的不同网络模型应用于肺结节轮廓识别,使用预处理后的LUNA16(Lung Nodule Analysis 16)数据集,引入批量归一化(Batch-Normalization)和残差网络进行优化训练。实验结果表明,模型能以较快的速度收敛,在测试集上的识别相似度为0.8796。
For information in English, please click here
Pulmonary nodule contour recognition in lung CT images based on U-Net network
Abstract:On the CT image of the lung, the pulmonary nodules are small and some of the pulmonary nodules are connected to the surrounding benign tissue. Since the gray value of the pulmonary nodules differs little from the gray value of the surrounding benign tissue, it makes the pulmonary nodule contour recognition very difficult. In order to complete the task of pulmonary nodule contour recognition, different network models based on U-Net were designed to be used for pulmonary nodule contour recognition. The pre-processed LUNA16 (Lung Nodule Analysis 16) data set was used to introduce batch normalization (Batch- Normalization) and residual network for optimal training. The experimental results show that the model can converge at a faster rate, and the recognition similarity on the test set is 0.8796.
Keywords: Pulmonary nodule contour recognition U-Net;Batch-Normalization;residual network
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于U-Net网络的肺部CT图像中的肺结节轮廓识别
评论
全部评论