小样本三维肺结节图像分割识别研究

Segmentation and Recognition of Small Sample Images of Three-Dimensional Pulmonary Nodules

为了进行小样本的肺结节检测和识别,减少医生的工作量及提高诊断的准确率,本文将V-net模型应用于肺结节图像的分割识别研究。首先对LUNA16数据集进行了翻转、重采样等一系列预处理,提取肺部当中的肺实质,然后通过V-Net模型对提取的肺实质进行肺结节检测。实验结果表明,V-net模型分割检测的肺结节像素准确率可以达到99%,召回率83%,对实际临床应用具有一定的参考意义。

In order to conduct small sample lung nodule detection and recognition,reduce the workload of doctors,and improve the accuracy of diagnosis,this paper applies the V-net model to the segmentation and recognition of lung nodule images.Firstly,a series of preprocessing processes such as flipping and resampling were performed on the LUNA16 dataset to extract lung parenchyma from the lungs.Then,the extracted lung parenchyma was detected for pulmonary nodules using the V-Net model.The experimental results show that the V-net model can achieve a pixel accuracy of 99%and a recall rate of 83%for segmentation and detection of pulmonary nodules,which has certain reference significance for practical clinical applications.