Detection And Recognition Of Lesions In CT Images Based On Deep Learning

The incidence of lung cancer is increasing and it has become one of the most dangerous malignant tumors to human health.Pulmonary nodules are a common lung disease.Although multiple pulmonary nodules are mostly benignning,some of them are at risk of becoming cancerous.Therefore,early detection and timely treatment of malignant nodules are particularly important.At present,in clinical practice,the nodules are mainly judged by the experts based on their personal experience and the observation of pulmonary CT images,so as to reach a diagnosis conclusion.Artificial detection is time-consuming and requires high professional quality of doctors,so it is difficult to ensure that all patients can get timely diagnosis and treatment.At the same time,there is a large subjective contingency in the artificial judgment,which will bring some differences in the detection results.In particular,some nodules have unclear boundaries,irregular shapes and great morphological changes in different scanning layers,which bring difficulties in the judgment and lead to patients missing the optimal treatment period.Computer aided diagnostic system can provide effective image interpretation and diagnostic advice for doctors.In recent years,the rapid development of deep learning technology provides a new technical means for medical image computer-aided diagnosis.The combination of medical knowledge and computer aided diagnosis system can greatly improve the diagnostic accuracy,and avoid the phenomenon of missed diagnosis and misdiagnosis,and ensure the timely detection of lesions and early treatment.In this paper,the automatic detection algorithm of pulmonary nodules based on deep learning is studied,and the algorithm of pulmonary nodules detection based on multi-scale detection algorithm and based on attention mechanism are proposed respectively.Through the comparative analysis of experimental results,the optimal model is determined to realize the high probability and accurate detection of pulmonary nodules.Finally,a pulmonary nodule detection system is established to provide technical support for the clinical application of the automatic detection algorithm for pulmonary nodule.In the research of lung nodule detection algorithm based on multi-scale strategy,three-dimensional convolutional neural network is used as the basic network to realize feature extraction of CT image sequence,fully consider the relationship between sections,and effectively utilize the spatial internal connection of lesions.Moreover,the model incorporates the Anchor mechanism,feature fusion,residual mechanism and multi-scale strategy.The Anchor mechanism can realize the accurate detection of pulmonary nodules of different sizes.The idea of feature fusion makes full use of the deep and shallow information.The residual mechanism enables the network to automatically select the appropriate network depth.The multi-scale mechanism effectively avoids the problem of missed diagnosis.The CPM(competition performance metric)score of this model reaches 0.8416.In the research of pulmonary nodule detection algorithm based on attention mechanism,the main idea is to pay attention to the key feature information by using attention mechanism and ignore the irrelevant feature information.In this paper,soft attention mechanism is mainly adopted to screen important information in threedimensional convolutional neural network by integrating channel attention and spatial attention,so as to enhance the feature expression of the network,improve the ability to detect nodules,and realize high probability and accurate detection of pulmonary nodules.In this paper,two attention mechanism models are built,and the structure of the model is adjusted by the coefficient k.Through comparative analysis of the experimental results and evaluation criteria,the model has a better ability to detect nodules when k is 1.When FPs/scan is 2,the model’s sensitivity reaches 0.9818.The CPM score of this model is as high as 0.8715,which is 1.31% higher than that of the model without adding attention.In terms of pulmonary nodule detection system,a pulmonary nodule detection system on the web page is built by using HTML language.At the same time,this paper adopts My SQL database to realize the storage function of data.The pulmonary nodule detection system realizes the functions of medical staff to input and view the patient’s information and diagnosis results.The Python based lung nodule detection algorithm is embedded into the system to realize the batch detection of lung CT images,which greatly improved the working efficiency of doctors.