FSJP-Net: Foreground and Shape Joint Perception Network for Glomerulus Detection

Qiuchi Han, Xiuxiu Hu, Pingsheng Chen, Siyu Xia

Abstract: Morphological changes in the glomerulus play a vital role in the diagnosis of kidney diseases. However, the detection of the glomerulus in the actual medical situation is challenging due to various factors such as lesions, tissue changes, and staining. These factors raise problems like high foreground-background similarity, blurred contours, and irregular shapes, thus pose difficulties for both physicians and Automatic computer detection. To address these challenges, we propose a foreground-aware feature extraction method, which is used to fully extract foreground information. Furthermore, we design the Foreground and Shape Joint Perception Network (FSJP-Net), a detection network that integrates object foreground information and shapes information, which improves the recall and precision of glomerular detection by fusing the extracted foreground and elliptical shape information from different feature extraction branches. The experiments demonstrate the effectiveness and superiority of our proposed method in detecting various categories of glomeruli.