Attention-DPU: Dual-path UNet with an attention mechanism for ultrasound image segmentation

Abstract

With the continuous advancement in computer vision, image segmentation has achieved fruitful results in many applications such as medical image processing. In recent years, UNet and Dual Path Network (DPN) have achieved promising results in medical image segmentation. UNet cannot effectively obtain new features and reuse features. Also, DPN cannot effectively transfer the contour information of shallow blocks to the subsequent deep blocks. This paper proposes a dual path U-shaped network with the attention mechanism (Attention-DPU); taking advantage of the two networks. In the proposed network, the ordinary convolutional layer is replaced by the micro block with a dual path. Also, the attention mechanism is adopted to improve the efficiency and accuracy of segmentation.