基于改进的3D U-Net的肺结节检测算法

Pulmonary Nodule Detection Algorithm Based on Improved 3D U-Net

针对传统方法对结节检测不全,且误检率较高的问题,提出了一种改进的3D U-Net肺结节检测算法。根据肺结节的形态大小特性,将残差网络引入3D U-Net网络框架中,并在3D U-Net的编码结构里引入了空洞卷积。利用改进的3D U-Net肺结节检测算法在中南民族大学认知科学实验室中完成一系列肺结节检测实验。实验结果表明,所提出的算法检测敏感度达到92.30%,并且降低了误诊率和漏检率。

In order to solve the problem of incomplete detection and high false detection rate of traditional methods,an improved 3D U-Net lung nodule detection algorithm is proposed.According to the shape and size characteristics of pulmonary nodules,the residual network is introduced into the 3D U-Net network framework,and the hole convolution is introduced into the 3D U-Net coding structure.The improved 3D U-Net pulmonary nodule detection algorithm was used to complete a series of pulmonary nodule detection experiments in the Cognitive Science Laboratory of South-central University for Nationalities.The experimental results show that the accuracy of the proposed algorithm is 92.30%,and the rate of misdiagnosis and missed detection is reduced.