A Study on Intrarenal Porous Segmentation of Endoscopic Images

The three-dimensional porous structure is a common scene in the field of medical surgery. However, the imaging resolution of the endoscope is low, and its scene has weak texture features and high noise, which leads it difficult for traditional algorithms to segment the porous structure. In this study, a new approach for efficient segmentation of intrarenal porous areas in endoscopic images is put forward. The proposed method first classifies the images into deep and shallow groups based on the statistical vectors derived from the color feature histogram. Then for each group, the improved U-Net learning strategy is used to extract the intrarenal porous areas at the pixel level, and its segmentation results could be accurately obtained. The effectiveness and accuracy of this work are evaluated on the data of the ureteroscopic holmium laser lithotripsy.