A Deep Learning Model Of Automatic Detection Of Pulmonary Nodues Based On Convolution Neural Networks (CNNs)

Recently,the haze again and seriously affect people's lives,but also a serious health hazard.Under the influence of haze,PM2.5 and other air pollution,the number of patients with lung cancer across the country and the world showed a trend of exponential growth.Lung cancer lesions called pulmonary nodules,it has the advantages of smaller size,shapes,and pleural adhesion of endometrial characteristics and bronchial vascular disturbance caused has certain difficulty in early diagnosis of lung cancer.At the same time,the area of small pulmonary nodules in the lungs for physicians,with the naked eye observation of the CT image,according to the existing knowledge and experience to identify lung nodules and lesions caused by misdiagnosis or missed diagnosis in the diagnosis of benign or malignant tumor.In the low dose CT scanning technology is widely used,the contradiction between image data explosion and artificial diagnosis of a serious shortage of power,the development of big data technology and data analysis of the coordination,are likely to lead to decrease the accuracy of the diagnosis of lung cancer.With the development and application of computer technology,in many large hospitals,doctors are aided by computer aided diagnosis of lung cancer.In the computer aided diagnosis of CAD,the classification of pulmonary nodules is realized by image preprocessing,segmentation,feature extraction and feature selection.It is the ultimate goal to improve the accuracy of classification of benign and malignant pulmonary nodules,and feature extraction is the key step.In this paper,the research status at home and abroad are studied,and the existing problems and solutions are put forward.According to the CT image of large data,the automatic diagnosis model of lung nodules based on convolution neural network is established.The main research work includes the following aspects:1.Resistible factors and 1 for complex computer aided diagnosis system algorithm and artificial interference generated by the image on the growth of lung parenchyma by simple pretreatment on CT images by region,through bilinear interpolation for sample storage.We use the sample to train the customized convolution neural network model to achieve the purpose of diagnosis of pulmonary nodules.This method can improve the accuracy and speed of classification of pulmonary nodules on the basis of avoiding complex algorithms such as feature extraction.2.Feature as the main classification of lung nodules,the extraction of features is an essential step.In the traditional methods,the feature extraction methods are based on the experience of artificial settings,including gray level,shape and texture and other low-level features,but these specific features have some limitations.In this paper,the original image is directly input into the convolutional neural network,and the key features are extracted from the hidden layer.In the existing methods,only the last output is used as the feature,and the hidden layer is ignored.Because the contribution rate of each feature is different,and the multi-layer learning may be lost in the last layer,this paper will get the final fusion feature by PCA dimension reduction.Although the fusion feature can not be described exactly,the classifier can get more accurate classification results.In this paper,the construction of convolutional neural network model,in the large sample data of the test certification,in reducing the complexity of the algorithm while improving the overall detection rate of pulmonary nodules,reducing the misdiagnosis rate and missed diagnosis rate.This provides a more accurate,effective and convenient method for the diagnosis of doctors,and has a positive effect on the early diagnosis and treatment of lung cancer.