Pulmonary Nodule Detection Technology Based On Deep Learning

Lung cancer is one of the nausea tumors with the highest morbidity and mortality in the world today.The diagnosis and treatment of early pulmonary nodules can effectively improve the survival rate of patients.The automatic detection technology of lung nodules is of great significance and value,and has always been one of the research hotspots in academia and industry.In recent years,with the rapid development of deep learning,convolutional neural networks have been widely used in the automatic detection of lung nodules,and have achieved great success.However,due to the complex three-dimensional structure of lung CT data and the diversity of lung nodules in shape and size,and a variety of similar structures,the correct detection of lung nodules is still a challenge.The large data set Luna16 is a commonly used public data set for lung nodule detection.The collection method is obtained by scanning the human lungs with CT instruments.It is a three-dimensional data structure that contains a series of multiple horizontal axial slices of the chest cavity.Due to the specificity of lung CT data collection,the difference in the scan rate of the collected images in the horizontal axis is small,but the difference in the vertical axis is large,because the number of slices included in each image will vary with the scanning machine,the scanning layer thickness and patient.In fact,the world physical scale corresponding to the lungs of each patient has a small difference,but the inconsistency of the scan rate on the vertical axis causes a large difference in the resolution of the obtained CT data of the lungs on the vertical axis.In order to deal with the impact of the inconsistency of the vertical axis resolution scale of lung CT data on lung nodule detection,the structure of the lung nodule detection network proposed in this paper includes two parts: a lung CT interpolation network based on voxel flow and three-dimensional convolution deep detection network.In the lung CT interpolation network,a convolutional neural network is first used to extract the voxel flow between adjacent slices,and then the interpolated lung slices are obtained by forward warping based on the extracted voxel flow and the required interpolation coefficients.After interpolation,the lung slice data with the same scale is realized,and then the nodules are detected through the three-dimensional convolution depth detection network.The experimental results show that the lung nodule detection method proposed in this paper can reduce the impact of the inconsistency of the vertical axis resolution scale of lung CT data on detection,and improve the accuracy of lung nodule detection.