Research And Implementation Of Auxiliary Detection Method Of Lung Nodules Based On YOLOv4

At present,in the worldwide ranking of malignant tumors,the incidence and mortality of lung cancer rank first.The early clinical manifestations of lung cancer are pulmonary nodules.Through the computer tomography(Computed Tomography,CT)of the patient’s chest,the nodules in the lungs can be detected,so as to achieve the purpose of early detection and early treatment.However,considering that the manifestations of lung nodules in the body are very diverse,under the traditional auxiliary diagnosis method,it is very easy to cause problems such as false detection and missed detection by relying on the doctors’ personal knowledge reserves to detect.In recent years,deep learning has achieved great success in image processing.Among them,convolutional neural network is the most widely used.It can perform features such as image feature extraction and classification,so that neural network can learn independently and improve model performance.Reduce human resource requirements.This article first preprocesses the CT image to filter out useless information,and obtains a better-displayed lung parenchymal image;then uses the neural network model to detect lung nodules in the lung parenchymal image;finally builds an auxiliary detection system for the convenience of doctors.The specific work is as follows:(1)Since the CT images collected from various medical institutions are quite different and cannot be operated uniformly,the images need to be standardized.Considering that lung CT contains a lot of useless information,segmenting the lung parenchyma and detecting lung nodules can improve the detection efficiency.However,the traditional segmentation algorithm performs poorly in segmenting the lung structure with small differences in features,so this paper proposes an improved drip segmentation algorithm,which can better segment the adhered lung structure without excessive segmentation,and segmentation Compared with traditional algorithms,the speed is improved by more than 40%.(2)After the comparative study,this thesis finally chose to use the YOLOv4 model for lung nodule detection,but this method still has a lot of room for improvement in detection accuracy.Improving the effect of model detection can start from two aspects: improving the network structure and optimizing parameters.First,redesign the adjusted network structure according to the imaging characteristics of the lung nodules;then design new network parameters after calculation according to the characteristics of the training data,and train the model to obtain the optimal weight;finally,compare several other classics horizontally Detection algorithm,the average detection accuracy of the optimized detection algorithm proposed in this thesis has reached 88.7%,which is generally improved by more than 8%.(3)In order to facilitate the use of doctors,an auxiliary detection system for lung nodules based on YOLOv4 is designed,which can greatly reduce the workload of doctors in clinical diagnosis.In this thesis,by improving the lung parenchyma segmentation algorithm and lung nodule detection algorithm,the accuracy and speed of lung nodule detection are effectively improved,and the auxiliary detection system is designed to improve the diagnosis efficiency of doctors.