Research On Pathological Image Classification Algorithm Of Non-Small Cell Lung Cancer Based On Improved ResNet

Lung cancer is the most common type of cancer among newly diagnosed cancer patients worldwide.Artificial intelligence has developed rapidly and has great prospects in medical image analysis.Deep learning in artificial intelligence has become an important auxiliary diagnostic tool for pathology experts.Therefore,a large number of researchers are committed to using deep learning methods to study medical image processing,which can improve the diagnostic efficiency while improving the diagnostic accuracy.The research in this thesis used convolutional neural networks applied to the automatic classification of lung cancer pathology images.The research objects of this thesis are the pathological images of lung adenocarcinoma,squamous cell carcinoma of lung in non-small cell lung cancer and benign lung pathological images.The research used the LC25000 dataset and achieved satisfactory results.The research work carried out in this thesis is as follows:(1)Used the Res Net residual network as the basic network model and used the Involution in the new Red Net network to improve and optimize the Res Net network,this improved method solved the problem of redundant information between channels and the small receptive field of traditional convolution kernels,and at the same time enhanced the ability of the network model to acquire and express global information.The accuracy of the network model in lung benign images,lung adenocarcinoma images and squamous cell carcinoma of lung images reached 100.00%,99.47% and 99.47%,respectively.And the accuracy of the entire dataset reached 99.47%.The experiments show that the improved model is effective in the non-small cell lung cancer pathological image classification task,and the classification accuracy is improved.(2)Res Ne Xt is improved on the basis of the Res Net and this thesis used it as the basic network model.The CBAM attention module is embedded in the Res Ne Xt to assign more weights to the useful features and improve the network to extract useful features.Used the DYRe LU module to replace the static Re LU in the initial layer which can automatically adjust the parameters according to different feature maps and improve the model performance.The accuracy of the network model in lung benign images,lung adenocarcinoma images and squamous cell carcinoma of lung images reached 100.00%,99.96% and 99.96%,respectively.And the accuracy of the entire dataset reached 99.96%.Experiments demonstrated the superiority of the improved model in the non-small cell lung cancer pathological image classification task.To sum up,the two parts of the research work in this paper can achieve good results in the task of non-small cell lung cancer pathological image classification,and the improved algorithm is innovative to a certain extent.The experimental results show that the two improved algorithms in this paper are effective with certain clinical reference significance.