基于深度学习的肺癌影像辅助诊断系统的设计与实现

Design and Implementation of Lung Cancer Image Auxiliary Diagnosis System Based on Deep Learning

肺癌对人类生命安全构成巨大的威胁,因此该文针对CT图像上出现的肺部结节特征不明显、形态大小不一且边界模糊不清的问题展开探讨,并给出基于A-VNET的系统针对肺部结节的划分方式。在该系统中设计了V-net框架,引入注意力机制,以突出某一局部位置的显著特点,从而改善模型的分割特性。在LUNA16数据集上做了分割实验,该方法 A-Vnet的F1分数比V-Net高2%,显著地提升了肺结节的分割精度。

Lung cancer poses a great threat to human life safety. Therefore, this paper discusses the problems of unclear features,different shapes and sizes, and blurred boundaries of lung nodules on CT images, and gives the classification method of lung nodules by A-VNET-based system. V-net framework is designed in the system, and Attention Mechanism is introduced to highlight the salient features of a local location, so as to improve the segmentation characteristics of the model. A segmentation experiment is conducted on LUNA16 dataset. The F1 score of A-Vnet is 2% higher than that of V-Net, which significantly improves the segmentation accuracy of lung nodules.