Research On Computer Aided Diagnosis Of Lung Cancer Based On Improved Canny Algorithm

Lung cancer is one of the malignant tumors with the fastest increasing morbidity and mortality in the world and the greatest threat to human health and life.Due to the appearance of early symptoms of lung cancer is not easy to detect,when physical discomfort or long-term cough,the hospital may have reached the advanced stage of lung cancer diagnosis,missed the best period of lung cancer treatment.Therefore,the early diagnosis and treatment of lung cancer has important research significance.Against this background,this paper conducted a research on computer-aided lung cancer diagnosis technology.The main research contents are as follows:(1)To analyze the pathological features and medical signs of pulmonary nodules as the theoretical basis for subsequent diagnosis and analysis,and to lay a solid foundation for the overall system framework.The international medical image standard DICOM and the corresponding DCM format were analyzed to pave the way for the subsequent image processing and preprocessing.The Canny algorithm was deeply studied and realized,and the results showed that the Canny algorithm had shortcomings and needed targeted improvement.(2)As for image segmentation and extraction,the Canny algorithm,which has a better effect on the segmentation of weak edges,is selected for the segmentation and extraction of lung nodules in view of the small difference in gray value between different tissues of lung CT images.Since the most important thing is to integrate into a complete system,the algorithm must be adaptive.In order to improve the selfadaptability,the enhanced Canny algorithm based on adaptive median filtering and iterative threshold selection algorithm was proposed in this paper.The high and low double thresholds and filter sizes that need to be manually selected in the Canny algorithm were automatically obtained.In addition,since pulmonary nodules may adhere to CT images,an open operation in morphological methods is proposed to separate the adhered nodules.Canny algorithm is prone to over-segmentation,so that the segmented regions are more than the actual regions,resulting in the false positive of the segmented pulmonary nodules.In this paper,a SVM-based false positive pulmonary nodules removal operation was proposed,and the pulmonary nodules were classified as false positive by using the image characteristics of pulmonary nodules,and finally the correct pulmonary nodules image in lung CT was obtained.After the overall improvement,the accuracy of real pulmonary nodules segmentation increased from 63.2% to 84.2% of the traditional Canny algorithm,and the improvement was relatively obvious.(3)In the analysis and diagnosis part,the most important way to determine whether a patient has lung cancer is to determine whether there are malignant pulmonary nodules in the lung.Because of the difference of malignant lung nodules and benign pulmonary nodules are not obvious,the characteristics of the small size of lung nodule images at the same time,this paper based on the Dense Net convolution neural network to classification of benign and malignant lung nodules are visible,with nodules itself has a small size and the characteristics of benign and malignant lung nodules difference is not obvious,so in this paper,the characteristics of reuse and the incoming of require smaller size Dense Net convolution neural network was improved.In view of the characteristics of the size of pulmonary nodules and the surrounding environment of pulmonary nodules,the Dense Net convolutional neural network was changed into a double-input single-output structure,and the features were fused by Concatenate method.In order to improve the overall classification effect,an improved Dense Net convolutional neural network based on Inception module structure was proposed.The Bottlenneck Layer structure and Inception module structure in Dense Net convolutional neural network were improved to further improve the diagnostic accuracy of the system.The diagnostic accuracy increased from 89.5% of the former Dense Net convolutional neural network to 94.7% of the latter.The effectiveness of the improvement was verified and finally demonstrated in the accuracy of diagnosis.(4)Finally,a lung cancer computer-aided diagnosis system was implemented on the Qt cross-platform graphical user interface application development framework,which mainly consists of four modules: patient information module,CT image processing module,benign and malignant pulmonary nodules classification module and CT image remark module.The CT images of patients can be analyzed and diagnosed to provide a preliminary diagnosis result for doctors' reference,so as to achieve the purpose of computer-aided diagnosis of lung cancer.