Can I read Lesions Segmentation of Medical Ultrasound Images with Boundary Enhancement Strategy on EtoBox?

Lesions Segmentation of Medical Ultrasound Images with Boundary Enhancement Strategy by Wentao Liao; Xuan Zhang; Wenbo Song; Xinglong Wu; Guoping Xu is a scholarly article available to read on EtoBox.

What is Lesions Segmentation of Medical Ultrasound Images with Boundary Enhancement Strategy about?

Segmentation plays an important role for ultrasound image analysis. However, it is still a challenging problem for automatic target segmentation on ultrasound image owing to the low resolution and low contrast with surrounding tissues or organs on ultrasound image. Previous work demonstrated that shape of the segmented object can provide priors knowledge to help estimate missing boundaries. In this work, we encode the boundary prior information into the loss function. Experiments are conducted with some networks such as CPFNet, U-Net, Fast-SCNN on the hemangioma ultrasound dataset and the breast cancer ultrasound dataset The best results are achieved by our proposed method on CPFNet with dice indices of 87.89% and 71.73%, respectively. The experimental results demonstrate that our boundary enhancement strategy can effectively improve the segmentation performance on ultrasound images without increasing the parameters of model. CCS CONCEPTS • Computing methodologies → Neural networks.

Author

Wentao Liao; Xuan Zhang; Wenbo Song; Xinglong Wu; Guoping Xu

Publisher

ACM

Published

2022

Language

EN