Automatic Identification System For Pulmonary Nodules Based On Support Vector Machine

Computer-aided diagnosis system is mainly to help doctors to carry out detection and diagnosis.At present,the main research direction of this system is the discrimination between benign and malignant cancer.In this paper,the identification of pulmonary nodules was taken as an example.The CT images of the lungs in the hospital physical examination were selected as the original data.Since only the lung parenchymal image part is needed in the CT image,the lung parenchyma is first divided as much as possible.Since the segmentation phenomenon occurs during the segmentation process,the lung parenchyma needs to be repaired at the edge.In the lung parenchyma image,we can obtain the ROI region.The quality of the recognition is affected by two factors,one is the choice of the kernel function,and the other is the parameter optimization algorithm.In order to achieve the best experimental results,this paper discards the traditional single kernel function method for the kernel function,and chooses the method of mixing two kernel functions.For the optimization parameter algorithm,the two algorithms are combined with each other.In this paper,in addition to the selection of traditional morphological features and grayscale features,texture features and invariant moment features are also selected.The main work of this paper is mainly as follows:(1)This article first supplements the corresponding medical knowledge,tells the overall design idea of the system,and also expounds the development of the auxiliary detection system at home and abroad.(2)The segmentation of the lung parenchyma is very large for the success or failure of the experiment,and the segmentation of the lung parenchyma.Firstly,the optimal threshold method is used to reduce the calculation time of the algorithm,so that the overlapping area of the gray value of the desired part and the unnecessary part in the image is reduced.Then,the connectivity is processed,and only the partial region of the lung parenchyma is preserved.It is then processed using a region growing algorithm.The processed image will have small noise and need to be denoised.The segmentation of the lung parenchyma will be over-segmented.In order not to affect the subsequent experiments,the image is edge-corrected,and a local minimum-value connection algorithm and connection algorithm are used.(3)The obtained lung parenchyma image was initially processed to obtain the candidate ROI region,and the ROI region was subjected to four aspects,and a total of 20 dimensions of feature extraction were performed.In the choice of support vector machine kernel function,we choose the mixed mode of polynomial kernel function and radial basis kernel function,and construct a mixed kernel function.The experimental results show that the mixed kernel function has better performance.(4)Another factor that affects the experimental results,the parameters of the kernel function.In this paper,the parameters are optimized to maximize the accuracy of recognition.The traditional genetic algorithm and the grid search method are combined,and the parameters are optimized by a combination method.Firstly,the characteristics of the probabilistic parameters of the genetic algorithm are used to perform a rough first search.Then use the grid search method to carry out detailed search and find the absolute optimal parameters.(5)This paper also designed and realized the automatic identification system of lung nodules,and detailed design and implementation of the system architecture,system modules and systm interface.In summary,the regional growth algorithm used in this paper is effective in extracting lung parenchyma.This paper focuses on the accuracy index,the hybrid kernel function composed of the radial basis kernel function and the polynomial kernel function,and the optimization algorithm after optimization optimizes the accuracy.