A Image Preprocessing Method Based on Visual Attention Mechanism for Deep Learning Detection of Pulmonary Nodules

Abstract

Computer aided detection (CAD) technology can automatically detect and identify suspicious
pulmonary nodules in CT images, so as to improve the detection efficiency of medical staff.For a large CT
image, if the deep learning technology is to be used to detect and identify the pulmonary nodules, the input
CT image should be preprocessed and processed into the size required by the deep learning network.
Therefore, it is very important to select the candidate areas of pulmonary nodules. In this paper, a
preprocessing method of pulmonary nodule image based on visual attention mechanism is proposed. After
the lung parenchyma is segmented from the chest CT image, combined with the characteristics of the
pulmonary nodules, the five features of direction, brightness, corner, edge and local entropy are used to
detect the significant areas of pulmonary nodules, and then the pulmonary nodules corresponding to these
significant areas are processed into the size required by the deep learning network. The experimental results
show that this method can be used to recommend the candidate areas of pulmonary nodules, which is of
great significance for the classification and recognition of pulmonary nodules using deep learning
technology.