Deep Convolutional Neural Network For Lung Nodule Automated Detection And Classification

Lung cancer is one of the diseases with the highest morbidity and mortality worldwide.Early diagnosis and treatment of lung cancer is an important means of reducing mortality.With the continuous development of medical imaging technology,CT image data has multiplied,but the number of experienced physicians is limited,which leads to the problem of explosive growth of medical imaging data and the serious shortage of manual diagnostic power.And with the extension of working hours,the physician Fatigue and distraction lead to missed detection and false detection.Therefore,computer-aided techniques are urgently needed to provide physicians with objective decision support and to assist physician diagnosis.Pulmonary nodules are early manifestation of lung cancer,lung nodule detection is a crucial step in diagnosing pulmonary cancer.The diagnosis of benign and malignant pulmonary nodules is very important for physicians to perform nodule staging and individualized treatment plans.In order to improve the detection rate and classification accuracy of lung nodules and better assist physicians in the early diagnosis of lung cancer,this paper proposes a new computer-aided detection and diagnosis method based on convolutional neural network.The hierarchical network structure of convolutional neural networks has powerful feature learning and feature expression capabilities,which can process massive medical image data and achieve excellent results in medical image analysis tasks.This paper focuses on the detection and classification of lung nodules based on deep convolutional neural networks.The major work of this paper includes two parts below:(1)Convolutional neural network with multi-level features fusion for lung nodule detection.Lung nodule detection plays an important part in the early diagnosis of lung cancer.Traditional lung nodule detection methods are complex and time-consuming,and the detection efficiency is low.In order to improve the efficiency of lung nodule detection,this paper uses deep learning to propose a novel computer-aided detection method,a CT image lung nodule detection method based on convolutional neural network multi-level feature fusion.Firstly,the pre-trained VGG16 model is used for feature extraction.Then,multi-level feature fusion method is used to fuse different levels of features to obtain a fusion feature map with richer semantic information,and then a full convolution network is used to generate nodule candidate boxes on the fusion feature map.Then,the deconvolution operation is performed on the fusion feature map,and the nodule recognition and the boxes regression are performed,and finally the pulmonary nodule is automatically detected.On the public dataset LIDC-IDRI,this method reaches high detection sensitivities of 85.6% and 90.3% at 1 and 4 false positives per scan,respectively.The proposed method not only can improve the detection rate of pulmonary nodules,but also can accurately locate the nodule position,and effectively assist the physician in early diagnosis,which is of great value for the diagnosis of lung cancer.(2)Fusing multi-dimensional convolution neural network for lung nodules classification.In order to solve the problem of low classification precision and high false positive in the classification task of lung nodules in CT image,a benign and malignant classification model of lung nodules based on weighted fusion multi-dimensional convolution neural network was proposed.The model contained two sub-models: a multi-scale dense convolutional network model based on two-dimensional images to capture more extensive nodule variation features and promote feature reuse,and the three-dimensional convolutional neural network model based on three-dimensional images to make full use of spatial context information of nodules.Firstly,2D and 3D CT images were used to train the sub-models.Secondly,the weights of the sub-models are calculated according to the classification errors,and then the weights were used to fuse the sub-models classification results.Finally,the more accurate classification results were obtained.The classification accuracy of the model was 92.25% and the AUC value was 98% on the public dataset.The experimental results show that the weighted fusion multi-dimensional model can effectively improve the classification performance of lung nodules.