Intelligent Diagnosis Of Lung Nodules Based On 3D Hybrid Features And Machine Learning

With the increasing air pollution,lung cancer has become a life-threatening disease.Early detection lesion areas has great significance for treatment.Traditional diagnostic methods are time-consuming and prone to misdiagnosis.Therefore,intelligent lung nodule diagnosis system based on machine learning appears.Computer aided diagnosis(CAD)faces a lot of difficult problems,such as large amount of test data,accurate segmentation of target area and low accuracy of pattern classification.To solve these problems,three diagnostic methods for pulmonary nodules are proposed.Specific research work is described as follows:(1)The research status of intelligent assistant diagnosis technology for pulmonary nodules is analyzed.The morphological characteristics,pathological causes and general diagnostic methods of pulmonary nodules are summarized.The general procedure of the assistant diagnosis system for pulmonary nodules is introduced.The basic theory of image preprocessing,image segmentation,feature extraction and pattern classification are expounded.(2)Lung nodule diagnostic method based on hybrid feature and support vector machine: Considering the noise in the collected CT images and the similarity of the gray level information between pulmonary nodules and blood vessels,the mean shift clustering algorithm is used for image preprocessing.Pulmonary nodule candidates are obtained by combining adaptive threshold and rolling algorithm.The hybrid features of gray level,shape and texture are extracted.LDA is used to select high-dimensional features.The constructed reduced sample set is input into the trained SVM model for nodule detection.A large number of simulation experiments show that the preprocessed image is more conducive to ROI segmentation,and the proposed nodule detection scheme effectively improves the classification accuracy.(3)Automatic pulmonary nodules diagnosis based on geometric active contour and logistic regression: due to the anisotropic diffusion algorithm does not take into account the characteristics of image gradient,the composite diffusion coefficient function is selected to preprocess the image.Geometric active contour algorithm is adopted for multi-target image segmentation.For the qualitative diagnosis of benign and malignant nodules,the three-dimensional nodules were reconstructed based on Kriging interpolation method and the 3D hybrid features were extracted.The constructed sample set was input into the PSO-LR model to classify benign and malignant tumors.A large number of simulation experiments show that the diagnosis method has advantages in image denoising,nodule candidate segmentation and diagnosis accuracy.(4)Pulmonary nodule diagnosis based on automatic image segmentation and random forest: an adaptive diffusion threshold de-noising algorithm is proposed based on the improved anisotropic diffusion algorithm.Fixed initial segmentation contour is adopted to avoid the manual definition trouble of seed points.A large number of simulation experiments show that the method has good generalization ability and improves the nodule diagnosis efficiency.Sum up,based on the massive CT images in LIDC-IDRI database,combined with image processing and data analysis,three diagnostic methods for pulmonary nodules are proposed in this paper,which reduce the time cost and computational complexity,greatly relieve the pattern classification pressure and improve nodule diagnosis efficiency.