Research On Lung CT Image-assisted Diagnosis Technology Based On Deep Learning

Lung nodule detection is the main method for early screening of lung cancer.Computeraided diagnosis(CAD)systems for lung nodule detection can automatically mark lung nodules in CT images with related algorithms,assist doctors in interpreting the information contained in CT images,and improve the efficiency and accuracy of doctors’ diagnosis.However,the CAD system for lung nodule detection still has the problem of high false positive rate and low accuracy rate.In order to solve the above problems,we have carried out research from three aspects: data enhancement,lung nodule detection and false positive reduction.The main research contents of this paper are as follows:(1)Aiming at the problem that the lack of medical image data makes it difficult to train deep learning algorithms,this paper proposes a generative data enhancement algorithm with dual discriminators.The algorithm adds a global discriminator based on the deep convolutional generation adversarial network(DCGAN),which can learn the background context information of the CT image while learning the data distribution of lung nodules,thereby synthesizing realistic nodules in the background of the CT image.This method can increase the detection accuracy of the Faster R-CNN detection model by 6.33%,and effectively alleviate the problem of small amount of medical image data and insufficient data diversity.(2)Aiming at the problem that the size of lung nodules is smaller than natural objects and the features are difficult to extract,this paper proposes a lung nodule detection algorithm based on improved Faster R-CNN.First of all,a deconvolution layer is added to the last layer of the VGG16 convolution layer to improve the resolution of the feature map.Then,the size and number of anchor frames were reset based on the size and volume characteristics of lung nodules.Finally,ROI Align is used to replace the ROI pooling operation to avoid positioning errors due to quantization operations during the feature mapping process.In addition,this article also conducts a comparative experiment with the original Faster R-CNN.Experiments show that the detection accuracy of the improved Faster R-CNN is 8.02% higher than that of the original Faster R-CNN,which verifies the effectiveness of the improved algorithm.(3)Aiming at the problem of high false positive rate in the detected candidate nodules,this paper proposes a 3D CNN classification network,and adds the SE module on the basis of the residual network.By weighting feature channels,SE module can enhance useful channel information and suppress useless channel information,and strengthen the feature expression ability of the model.Experimental results show that the 3D CNN network can effectively identify false positive nodules and reduce the number of false positives nodules.In summary,this paper has conducted an in-depth study on the detection of lung nodules in medical CT images.The research content includes generative data enhancement,lung nodule detection based on Faster R-CNN,and 3D CNN false positive reduction.The above research content has a positive role in promoting the development of lung nodule detection in medical CT images.