Research On Lung Nodule Detection And Classification Algorithm Based On Artificial Neural Network

According to the latest statistical report of China’s cancer center,the mortality rate of lung cancer is far ahead of other cancers,which seriously threatens human life and health.As an early manifestation of lung cancer,lung nodules are detected early and play a very important role in their quasi-certainty,so how to accurately locate lung nodules and accurately classify benign and malignant has become a hot spot in the current medical research field.Clinically,the diagnosis of lung nodules is mainly done by Computer Tomography(CT)images,but CT images have problems such as low contrast and strong interference by surrounding tissues,which makes the detection and classification of nodules difficult.At present,the detection and classification of lung nodules are mainly based on the doctor’s many years of clinical experience,which may cause the doctor to misdiagnose in the diagnosis process due to some factors.Based on the above problems,this paper proposes a pulmonary nodule detection method based on structural similarity and a DNN-based classification method for benign and malignant pulmonary nodules,which can improve the accuracy of lung nodule detection and classification,which can effectively help doctors reduce the misdiagnosis rate and improve the diagnosis efficiency.In this paper,lung CT images were taken as the research object to study the optimal segmentation of PCNN,nodular area detection,feature extraction,and benign and malignant classification.The main research work of the thesis is as follows:(1)This paper proposes a method to select the optimal times of iterations of PCNN segmentation.The relationship between the output of segmentation iteration and the grayscale of image is analyzed.Then,the gray features of image are extracted by wavelet transform and maximum wavelet coefficient pooling method,and DNN is used to accurately predict the times of optimal segmentation iteration.(2)A nodule detection method based on structural similarity is proposed,by analyzing the grayscale statistical characteristics of the nodule region,which transforms the nodule detection problem into the structural similarity problem between the region of interest and all annotated templates,and determines the nodule region by calculating the average of the structural similarity with all templates.(3)Aiming at the problem that the traditional feature type is single,which is easy to lead to low classification accuracy of lung nodules,a feature extraction method that integrates traditional features and depth features is proposed.By extracting the shape,gray level,texture,and depth features of the pulmonary nodule parenchymal area,all features were fused by linear concatenation.(4)Benign and malignant classification of lung nodules based on deep neural network.Through a large number of experiments,the optimal network framework was found to classify benign and malignant lung nodules.Compared with other classification algorithms,it has certain advantages in terms of accuracy,sensitivity and specificity,indicating that the classification algorithm proposed in this paper has certain application value in helping doctors diagnose.