EAU-Net++: An Efficient Segmentation Network for Ultrasound Vessel Images

Vessel segmentation on ultrasound images is an important tool for surgical planning and execution, but proves to be a challenging task due to noise and speckle. In order to achieve accurate surgical treatment and diagnosis, 386 ultrasound images from 10 subjects were manually marked. The collected data was used for data expansion through rotation, flipping, histogram equalization and least squares image transformation methods to alleviate the adverse effect of small data set. For the segmentation of ultrasound blood vessel images, this paper proposes the EAU-Net++ network. On the basis of U-Net++, EfficientNet is employed as an encoder. An attention mechanism is introduced to furtherly improve the segmentation accuracy. On the unified dataset, ablation experiments and comparative experiments are carried out to verify the proposed model. Experimental results show that EAU-Net++ can effectively achieve blood vessel image segmentation.