Research On The Segmentation And Classification Method Of Central Type Lung Cancer Based On Neural Network

Lung cancer is a prevalent disease with the leading prevalence and mortality rate in the world.Lung nodules,as a manifestation of lung cancer,can largely reduce the risk of death and protect the lives of lung cancer patients by performing computed tomography（CT）scans of suspected lung nodules.However,as medical equipment continues to develop and be widely used,the accuracy and number of tomographic images are gradually improving,and as a result,the demand for lung imaging to detect pulmonary nodules is increasing dramatically,which causes a great deal of time and effort for physicians to perform a large number of readings,resulting in great stress and burden and the risk of misdiagnosis or missed diagnoses.Therefore,this thesis studies lung cancer image maps,focusing on segmentation and benign-malignant differentiation of central lung cancer,and puts forward a segmentation algorithm based on Internal-Activity-Changed FCMSPCNN,a segmentation algorithm based on an im-proved U-net network model,and an improved convolutional neural network classification algorithm to farther strengthen the accuracy of segmentation and classification.The details of the study are as follows:（1）To address the problem of high computing complexity of the classical PCNN model,an improved FC-MSPCNN（IAC-FCMSPCNN）model is proposed,which further optimize and enhance the synaptic weight matrix W_ijkl,link strength parametersβand dynamic threshold E_ij[n],add a new set of modulation parameters to better control the size of the dynamic threshold,reduce the number of model iterations and decrease the model accuracy complexity.Then the experimental validation was performed using the maps from the PET-CT lung cancer atlas,and the assessment indexes such as accuracy and recall rate were improved by this method compared with other segmentation methods of the same type.Therefore,this method has better segmentation effect and is more suitable for clinical medical image segmentation,which provides a new idea for the development of lung nodule segmentation algorithm.（2）Due to the influence of related tissues around lung nodules on the selection process of lung nodules,an enhanced U-net network model is present in this thesis for the segmentation of central type lung cancer.By adding a channel attention module and a spatial attention module to the U-net,the ability of the network to extract information from local features is improved,and then the features obtained from the two attention modules are later fused to enhance the correlation between the features at each level,thus improving the segmentation effect of the image,and the experimental results show that the segmentation accuracy of the method reaches 0.914 and the recall rate reaches 0.865.In addition,this thesis also designs relevant experiments to compare with FCN＿8S,Seg Net and U-Net networks,which have better results and provide more accurate and convenient reference for clinicians’diagnosis,and have great supplementary effect and practical utility for the diagnosis and treatment of early stages of lung cancer patients.（3）Based on the lung nodules segmented by the improved U-net model,this article puts forward an modified Convolutional Neural Network model for the determining of benign and malignant of central type lung cancer,which uses convolutional layers instead of pooling layers in traditional convolutional neural networks to achieve the purpose of increasing the depth of the model,and then classifies the segmentation results as benign and malignant,and obtained better results,in which the accuracy rate got 0.915,the recall rate got 0.898 and the AUC reached 0.92.In addition,this thesis also compared with classical classification models such as VGG16,Res Net,Deep Lung and 3DMVCNN,and the experimental results showed that the present classification model can improve the classification performance of lung nodules with good classification effect.Therefore,the approach proposed in this thesis regarding the segmentation and classification of central lung cancer has achieved better experimental results from both subjective and objective aspects compared with other algorithms of the same type,which provides a new idea for physicians to segment and classify central lung cancer,with certain practical significance and application value.