Research On Feature Fusion And Recognition Algorithm Of Lung Nodules Based On Deep Learning

Cancer seriously endangers human life and health,among which lung cancer has always been the leading cause of cancer death.An effective way to reduce lung cancer mortality is early screening.The Lung Computer Aided Diagnosis(CAD)system can provide radiologists with an auxiliary third-party opinion to improve the detection efficiency and the identification accuracy of lung nodule..The selection and fusion of nodule features and the improvement of the recognition algorithm are crucial to the improvement of the performance of the lung CAD system.Based on the Hybrid Deep Learning(HDL)model,this paper mainly focuses on three key algorithms of feature extraction,feature fusion and nodule recognition in lung CAD system.(1)It is proposed to embed the Convolutional Block Attention Module(CBAM)into the Convolutional Neural Network(CNN)models VGG16 and VGG19 to construct Attention-embedded VGG16(AE-VGG16)and Attention-embedded VGG19(AE-VGG19).To improve the feature extraction ability of the network and makes the network pay more attention to the crucial feature information in the nodule description.Then,the parameter-based transfer learning method is used,and the parameters of the pre-trained model on Image Net are used as initialization parameters,and the weights are retrained with the candidate nodule images after preprocessing,which reduces the training cost and supplements the information loss in the target domain.(2)Principal Component Analysis(PCA)is used to reduce the dimension of the extracted deep features to retain the original feature information to the greatest extent.Then,Canonical Correlation Analysis(CCA)is used to fuse the dimensionality-reduced features,and only the feature vectors with strong correlation between the two sets of feature vectors are selected for fusion.The fused features have strong feature expression ability and low-dimensional characteristics,which not only reduces the amount of calculation,but also helps to improve the subsequent recognition effect.(3)Multiple Kernel Learning Support Vector Machine based on the Improved Particle Swarm Optimization(MKL-SVM-IPSO)is proposed for lung nodule recognition.In order to solve the problem of slow model training and easy to fall into local optimum,the Multiple Kernel Learning Support Vector Machine(MKL-SVM)composed of radial basis kernel function and polynomial kernel function is used.In order to speed up the training and convergence of the model,a Particle Swarm Optimization(PSO)with adaptive inertia weight is introduced,and the parameters are quickly optimized according to the fitness value.The dynamic learning factor is further adopted to adjust the self-learning ability and collective learning ability of particles.The experiment selects the public data set LUNA16 for training and testing.The results show that the accuracy of lung nodule recognition of the proposed lung CAD system can reach 99.56%,and the sensitivity and F1-score can reach 99.3% and 0.9965,respectively,which can reduce the possibility of false and missed nodule detection.