Study On Classification Method For Benign And Malignant Lung Nodules Based On Deep Learning

Lung cancer is a major disease that threatens human life and health,and its mortality and incidence rate are among the top two cancers in the world.Early symptoms of lung cancer are not obvious,which makes it easy to miss the best treatment and recovery time.Therefore,early lung cancer screening is crucial.National lung cancer screening trials have shown that early lung cancer screening and treatment with low dose CT can help reduce lung cancer mortality.However,with the popularity of early lung cancer screening,a large number of CT scans are available and the workload of doctors has increased dramatically due to the huge amount of information.In addition,differences in clinical experience have led to individual differences among doctors.Therefore,missed diagnosis and misdiagnosis are prone to occur during the diagnosis of lung cancer.With the development of computer vision and medical image processing technology,combined with the knowledge of imaging,physiological and biochemical,the computer-aided diagnosis system has been proposed,which is called “the third eye”of doctors.The system can help improve the sensitivity and specificity of lung cancer diagnosis.However,high false-positive and false-negative rates remain a major challenge for the automatic recognition system of benign and malignant lung nodules.To address this issue,the open-source lung image database consortium and image database resource initiative database were used as the research object.This study proposes a convolutional neural network model based on three-dimensional multiview and attention mechanism.It was applied to the classification of benign and malignant lung nodules with improved sensitivity and specificity.The framework consists of the following four parts:The first part is data processing.The “gold standard” data were obtained by cleaning the data.To solve the problem of inconsistent resolution and image contrast between samples,the CT images are resampled and normalized.For the imbalance of positive and negative samples,three data balance methods are compared and discussed.It’s concluded that the random translation method is more effective in solving the imbalance of benign and malignant lung nodules.In the data enhancement stage,random translation,rotation,and flip are used to expand the data volume.Finally,the pathological information is matched with the doctor annotations to provide data support for consistency analysis.The second part is to construct a benign and malignant classification model for lung nodules.Based on the characteristics of strong spatial heterogeneity and variable shapes of lung nodules,a three-dimensional multi-view framework was established by imitating the diagnostic process of doctors for suspicious lung nodules.Due to the differences of different views,a squeeze-and-excitation algorithm is introduced in the feature fusion stage.Finally,a three-dimensional multi-view squeeze-and-excitation convolutional neural network is trained to obtain a benign and malignant classification model for lung nodules.The third part is the results and analysis.The performance of the benign and malignant lung nodule classification model was verified by model evaluation metrics and consistency analysis.In the model evaluation stage,the multi-view model showed better classification results when compared with the baseline three-dimensional single-view convolutional neural network model.After learning the differences of each view,the three-dimensional multi view squeeze-and-excitation convolutional neural network model achieves better classification performance.The sensitivity of the binary classification model of benign and malignant lung nodules was 98.60%,the specificity was 90.00%,and the accuracy of the ternary classification model for benign and malignant lung nodules was 87.76%.Compared with the existing research methods,the method proposed in this study can more fully learn the spatial characteristics of lung nodules,and the classification performance is better.In the consistency analysis stage,the consistency between model prediction and pathological diagnosis is significantly higher than that between doctor annotations and pathological diagnosis,indicating that the method proposed in the study has certain application value and significance at the clinical level.The fourth part is the construction of benign and malignant classification system for lung nodules.The above method is programmed in the classification system of benign and malignant lung nodules based on deep learning by using Python language.The automatic classification convolutional neural network model based on threedimensional multi-view and attention mechanism can effectively identify benign and malignant lung nodules.The model is stable and reliable,which can be helpful for early screening of lung cancer.