Endoscopic Segmentation of Kidney Stone based on Transfer Learning

• DOI: 10.23919/ccc52363.2021.9550652
• Publication Date: 2021-10-07T04:24:31Z
• Authors: Rui Li; Yu Zhao; Yu Dai; Gongping Chen; Jianxun Zhang; Liang Cui; Xiaotao Yin; Xiaofeng Gao; Ling Li
• Keywords: Cross entropy

Endoscopic stone segmentation is of great significance in the comprehensive diagnosis and surgical planning ureteroscopic lithotripsy. Due to quality problems endoscopic imaging, such as artifact, highlight, reflection, contrast imbalance blur, it a challenge segment kidney fragments different shapes. In this study, improved U-Net model was used extract fragment area at pixel level, its contour information could be accurately obtained. The VGG16 network with strong portability encoder semantic multiple feature layers, up-sampling realized by using transposal convolution gradually restore details. experiment, DCE Loss combining Dice Cross Entropy adopted function model. experimental results show that has higher accuracy for from images, MPA, MIoU F1 score 96.44%, 97.62% 97.03% respectively. modified 2.25% than Deeplabv3 + model, 14.24% standard