Intelligent Analysis Of Pathological Characteristics Of Pulmonary Nodules CT Images And Key Techniques Of Information Retrieval Based On Image Feature

With the environment becomes worse and worse,and bad habits of people living,the incidence of respiratory diseases increased year by year,especially lung cancer morbidity and mortality ranked first in all cancer,so the early prevention and treatment of lung cancer has become particularly important.Computer-aided diagnosis has become an important research direction in the field of medical information by using computer to process and analyze images.Computer-aided diagnosis has achieved a high level of diagnosis in some areas,even comparable to human experts,but in the diagnosis of lung disease,there are still low accuracy,lack of diagnosis and other defects.To tackle those problems and face the challenges,we use the deep learning technology by the patient's lung CT image data,identified the patient's pulmonary nodules(size <20mm)for benign and malignant,intelligently analyzed the patient's lung cancer(size?20mm)pathological type,and designed a high-dimensional retrieval system for medical images with feature extraction based on deep learning,improved the accuracy of computer-aided diagnosis,to help clinicians to make faster and more accurate diagnosis of the disease.The research work of this paper mainly includes the following points:1.Designed and studied a method for the identification of benign and malignant CT images of lung lesions based on deep learning system by the patient's image data.The basic process includes:(1)the design of the deep learning network system;(2)data collection and pretreatment of patient images and pathologic results;(3)the training of Deep Learning Convolutional Neural Network Model;(4)and identification of patients with pulmonary nodules CT images of benign and malignant by using CNN model.In this study,a total of 921 cases of pulmonary nodules were collected from three data centers.The convolutional neural network was designed for the small size image data.Analyzed the accuracy of the deep learning network model for the identification of nodules with different sizes and types of nodules.Analyzed the accuracy of the deep learning network for the identification of benign and malignant nodule of different data centers.2.Studied the method of intelligent analysis for CT images of non-small cell lung cancer based on deep learning system,which can analyze the pathological characteristics of non-small cell lung cancer by patient image data.The basic process includes:(1)the design of deep learning network system;(2)data collection and pretreatment of patient images and pathologic results;(3)the training of Deep Learning Convolutional Neural Network Model by using transfer learning;(4)and the analysis of pathological characteristics of non-small cell lung cancer by convolution neural network model.3.Designed a content based high-dimensional medical image retrieval system.The system was used to extract the lesion,extract and select features of the lesion,and construct high-dimensional feature database using VA-Trie index structure.With the VA-Trie index structure,the retrieval system can solve the "dimension disaster" problem,and the index performance can be effectively guaranteed.4.Studied the image feature extraction method for solitary pulmonary nodules by using deep learning system.The method was combined with the content-based medical image high-dimensional retrieval system which was constructed with VA-Trie index structure.The results showed that feature extraction of the image by deep learning system can get higher retrieval accuracy than that of the artificial feature extraction.