Improved Cascade R-CNN for Medical Images of Pulmonary Nodules Detection Combining Dilated HRNet

Abstract

Using Computer-aided Diagnostic (CAD) to analyze medical images is currently a focused area, and deep learning is widely used in the detection of pulmonary nodules in medical imaging. Current detection algorithms are effective in detecting large pulmonary nodules, but their detection effect on small nodules and micro-nodules is not ideal. In order to solve this problem, this paper uses high-resolution network (HRNet) as the backbone network of Cascade R-CNN to improve its detection accuracy on small targets. HRNet can preserve the information of small target nodules in the feature map with high resolution and obtain a finegrained feature map for the detection task. This paper also combines dilated convolution with HRNet and proposes an improved HRNet named dilated HRNet. Experiments on the LIDC-IDRI dataset show that the improved Cascade R-CNN increases the detection accuracy of pulmonary nodules, especially on small nodules.