Research Of Pulmonary Nodules Detection And Semantic Feature Classification Method Based On CT Image

Due to the effects of smoking,genetics,oil fume,and the environment,and with the aging of China’s population,the incidence and mortality of lung cancer ranks first in China’s common malignant tumors,and the total number of lung cancer patients continues to rise.According to medical data,the 5-year survival rate of early lung cancer is higher than that of advanced lung cancer.The main ways to improve the prognosis are early detection,early diagnosis and early treatment.Low-dose chest CT is an internationally recognized effective method,but currently it is facing a large number of chest CT screening populations and the pressure of imaging physicians,and medical resources do not match the needs of patients.And with the application of thin-layer low-dose CT,such as the improvement of the display rate of small nodules,the quantitative measurement of nodules,the doubling of the number of images,etc.,make it difficult to read.In addition,the current serious shortage of medical resources and medical staff in our country is exhausting the physical and mental health of imaging physicians,which increases the risk of missed diagnosis and misdiagnosis.In recent years,core technologies such as big data technology and deep learning have flourished in the medical field.Physicians all use computers to assist in the diagnosis of lung cancer.Deep learning-based CAD has now covered clinical stages such as lesion detection,pathological diagnosis,and postoperative prediction.In early research,the automatic segmentation of lung nodules was completed with the help of CAD,and the suspected lung nodules were quickly and accurately located.Algorithmic models obtained from large data sets can avoid subjective biases,although some model screening results included some false Positive nodules,but significantly reduce the occurrence of false negatives,significantly reducing the workload of the imaging physician.In addition,it can also extract the location and morphological information of lung nodules,and further provide some decision-making opinions such as classification of lung nodules(solid,sub-solid and calcification)and tumor benign and malignant grades for doctors’ reference.This article analyzes the current status of research at home and abroad,and proposes solutions to existing problems.For the CT images of big data,an automatic lung nodule diagnosis model based on convolutional neural network is established.The main research work includes the following aspects:1.As the main detection and classification premise of lung nodules,feature extraction is an essential step.In the current deep methods,feature extraction methods use residual networks or densely connected networks for feature extraction,so that features are missing or duplicated,thereby reducing the accuracy of the experiment.This paper uses dual paths after preprocessing lung CT images.The network performs feature extraction to ensure that features are not missed and not duplicated,and the dual-path network is embedded in U-NET to implement feature cascading and generate candidate frames of different scales to ensure detection accuracy.2.The current two-stage method of lung nodule detection,that is,inserting a pooling layer to unify the candidate box scale when generating candidate nodules,and then performing feature extraction and final classification,so that the detection time is long.This paper embeds the feature extraction network into a class U-Net network extracts candidate frames,uses a convolution layer to map the feature maps divided into nine regions to corresponding position-sensitive feature maps,and performs pooling operations on the nine feature maps to generate corresponding position-sensitive pooling Layer,and then perform a voting operation on the pooled feature map,and divide the candidate box into nodules and non-nodules according to the score,to achieve end-to-end automatic detection.3.At present,most of the lung nodule classification methods are only for benign and malignant classification,but in the actual diagnosis of the physician,a variety of semantic features will be integrated to determine the benign and malignant.The scoring simulates the diagnosis process of the physician,and provides a more accurate,effective and convenient method for the diagnosis of the physician,which has a positive effect on the early diagnosis and treatment of lung cancer.