An End-to-end Image Feature Representation Model of Pulmonary Nodules

Abstract

Lung cancer is a cancer with a high mortality rate. If lung cancer can be detected early, the mortality rate can be greatly reduced. Lung nodule detection based on CT or MRI equipment is a common method to detect early lung cancer. Computer vision technology is widely used for image processing and classification of pulmonary nodules, but because the distinction between pulmonary nodule areas and surrounding non-nodule areas is not obvious, general image processing methods can only extract the superficial features of the image in pulmonary nodules. The detection accuracy cannot be further improved. In this paper, we propose an end-to-end model for constructing feature representations for lung nodule image classification based on local and global features. First, local plaque regions are selected and associated with relatively intact tissue, and then local and global features are extracted from each region. Deep models represent features that implement high-level abstract representations that describe image objects. The test results on standard datasets show that the method proposed in this paper has advantages on some evaluation metrics.