Research On Detection Algorithm Of Lung Nodules Based On Attention Model

Based on computed tomography(CT),the detection of lung diseases can be realized.CT is widely used in the detection of lung diseases and the analysis of disease progression due to its advantages of rapid film extraction,clear imaging and low cost.At present,the main method to evaluate lung diseases is computer aided diagnosis(CAD)system,but the traditional lung CAD system mostly relies on manual extraction of features,which is not only time-consuming and laborious,but also lacks standardization and uniformity.Lung CAD system based on deep learning has become the primary research target of researchers because of its advantages of fast detection speed and high accuracy.Convolutional neural network(CNN)in deep learning can achieve precise feature extraction,detection and recognition by intelligently learning samples.However,in medical image processing,deep learn-based lung CAD system still has problems of insufficient sample data with doctor’s notes and low detection accuracy of pulmonary nodules.The main research content of this paper is as follows:Data enhancement of pulmonary nodules was completed based on CT-GAN technology.Since the use of Dropout in the original CT-GAN would cause overfitting problems,we introduced Dropblock into CT-GAN to solve the above problems and improve the quality of generated nodules in a regular way of discarking information.At the same time,c T-GAN technology can quickly and accurately generate pulmonary nodules and corresponding labeling,and the way to expand the data set is more rapid and convenient.The experimental results showed that the quality of pulmonary nodules generated by improved CT-GAN was better.Due to study in depth detection under the framework of the sampling process will produce lung nodule characteristics of leakage problems,therefore in YOLO-the V4 framework introduced in coordinate mechanism of attention(CA),to capture the location aware of lung nodules,directionality,and across the channel information,help the model more accurate detection of lung nodules is interested in the area,thereby decrease the residual of nodules and checked by mistake.Experimental results show that the detection accuracy of the proposed model is better.