Research On Pulmonary Nodule Detection Based On Convolutional Neural Networks

Lung cancer is one of the serious diseaseswhich can compromise health and even become life threatening.Lung cancer potentially manifests itself as malignant pulmonary nodule in an early stage.Early detection of this disease can significantly improve the five-year survival rate of lung cancer patients.Therefore,accurate identification of pulmonary nodules in lung CT images is an important part for the prevention and treatment of lung cancer.The traditional pulmonary nodules detection algorithms rely on accurate ROI region extraction and exhaustive feature extraction.Inaccurate segmentation algorithms and feature extraction will directly affect the recognition results.In view of the deficiency of the traditional pulmonary nodules recognition algorithm,this paper focuses on the detection and recognition method of pulmonary nodules based on convolution neural network.The main contribution of this paper are summarized as follows:(1)In order to obtain accurate and uniform training samples of pulmonary nodules,The sample dataset was built from the LIDC dataset in this paper.Then,we design a method by using XML file to extract pulmonary nodules and non-pulmonary nodules from the pre-processed lung CT images,so that we can get unified data samples.(2)In order to compare the difference between the traditional pulmonary nodule detection and recognition algorithm and the pulmonary nodule detection and recognition algorithm based on CNN model,the key technology and implementation scheme of the traditional algorithm were studied in depth,where focuses on the accurate extraction of ROI and the detailed feature extraction process.In view of the problems of holes and roughness of the binarized lung CT images,the morphological operations,ball algorithm and other methods were used in this paper,so that we can obtain a complete lung parenchyma region.The characteristics of pulmonary nodules are extracted from three aspects:intensity,morphology and texture.Finally,the support vector machine is used to detect the pulmonary nodules.(3)In order to modify the deficient of the traditional pulmonary nodules detection algorithm,the convolution nerve network is introduced to identify pulmonary nodules.This model input the original ROI to extract features by using the network,which does not rely on accurate segmentation of pulmonary nodules and feature extraction.(4)In view of the shortcomings of the original CNN model,such as low stability when the number of iterations is low,a pulmonary nodules detection and recognition algorithm based on an improved CNN model is proposed in this paper.Firstly,the Dropout strategy is introduced and then the ReLU activation function is used due to its fast convergence speed.Finally,the CNN-RF model and the CNN-SVM model are proposed by incorporating the traditional classifier random forest and support vector machine.The experimental results show that the CNN-RF model is superior to the original CNN model,and it can maintain good stability even when the number of iterations is low.At the same time,compared with the current popular pulmonary nodules classification algorithms,this model has a high accuracy and low missed rate.(5)Based on the above research results,the pulmonary nodules detection system is developed by using Python and OpenCV.The system is mainly used to improve the eff-iciency of doctors and assist doctors to determine and identify pulmonary nodules.We focuses on the study of detecting and identifying pulmonary nodules based on convolution neural network,and introduces a variety of strategies to improve CNN model in this paper.Experimental results show that the CNN-RF algorithm proposed in this paper has high accuracy and low leakage rate,which can effectively assist doctors to identify pulmonary nodules.