基于深度学习的医学图像肺结节检测

Pulmonary nodule detection of medical image based on deep learning

为了解决传统计算机辅助诊断系统对肺CT图像结节检出率低、假阳性高的问题,提出一种基于深度学习的肺部CT图像结节检测模型.根据肺部CT图像的三维本质特性,首先采用3D Faster R-CNN提取特征,进行候选结节的检测;然后再利用3D卷积神经网络进行假阳性结节去除.该方法在LUNA16数据集上进行了实验,采用国际医学影像领域通用的FROC方法进行评价,统计在不同假阳比例下的敏感性指标,平均FROC数值为82.8%,相比于传统的诊疗方法识别率有显著提升.该模型可用于辅助医生进行肺结节诊断,具有一定的临床应用价值.

In order to solve the problem of low detection rate and false positive of lung CT image nodules in traditional computer-aided diagnosis system,a model for pulmonary nodule detection of CT images based on deep learning was proposed.According to the three-dimensional nature of the CT image of the lung,the 3 D Faster R-CNN is used to extract the features and the candidate nodules are detected.Then the 3 D convolutional neural network is used to remove the false positive nodules.The method is tested on the LUNA16 dataset,and is evaluated by the FROC method commonly used in the international medical imaging field.The sensitivity index under different false-positive ratios is counted,and the average FROC value is 82.8%,which is compared with the traditional diagnosis and treatment method.The rate has improved significantly.This model can be used to assist doctors in the diagnosis of pulmonary nodules,and has certain clinical application value.