基于改进LeNet-5的肺结节检测方法研究

Pulmonary Nodule Detection Method Based on Improved LeNet-5

目前,肺部CT图像的筛查工作大多数是由影像科专家人工读片完成,随着医学影像服务需求的不断增长,这种繁重的人工阅片工作无疑会成为医生的沉重负担.因此,本文提出一种基于改进LeNet-5的肺结节检测方法.该方法基于LIDC-IDRI肺癌检测数据集,采用卷积核大小为3、步幅为2的卷积层取代传统LeNet模型中原有的池化层,并且在每层Activation Function之前加入批量归一化层.实验结果表明,与传统的LeNet模型相比,该模型在肺结节识别上有更好的效果,从而进一步辅助影像科医生进行诊断,降低工作强度.

At present,most of the screening work of lung CT image is done by the experts of imaging department.With the increasing demand of medical image service,this heavy manual reading work will undoubtedly become a heavy burden for doctors.Therefore,a lung nodule detection method based on the improved LeNet-5 is proposed in this paper.This method is based on LIDC-IDRI lung cancer detection data set,using a convolution layer with a convolution kernel size of 3 and a stride of 2 to replace the original pooling layer in the traditional LeNet model,and adding batch normalization before each Activation Function.The experimental results show that compared with the traditional LeNet model,the model has a better effect on lung nodule recognition,thereby further assisting the imaging doctor in the diagnosis and reducing the work intensity.