AAU-net: An Adaptive Attention U-net for Breast Lesions Segmentation in Ultrasound Images

TL;DR

An adaptive attention U-net to segment breast lesions automatically and stably from ultrasound images is developed to replace the traditional convolution operation and robustness analysis and external experiments demonstrate that the proposed AAU-net has better generalization performance in the breast lesion segmentation.

Abstract

Various deep learning methods have been proposed to segment breast lesion from ultrasound images. However, similar intensity distributions, variable tumor morphology and blurred boundaries present challenges for breast lesions segmentation, especially for malignant tumors with irregular shapes. Considering the complexity of ultrasound images, we develop an adaptive attention U-net (AAU-net) to segment breast lesions automatically and stably from ultrasound images. Specifically, we introduce a hybrid adaptive attention module, which mainly consists of a channel self-attention block and a spatial self-attention block, to replace the traditional convolution operation. Compared with the conventional convolution operation, the design of the hybrid adaptive attention module can help us capture more features under different receptive fields. Different from existing attention mechanisms, the hybrid adaptive attention module can guide the network to adaptively select more robust representation in channel and space dimensions to cope with more complex breast lesions segmentation. Extensive experiments with several state-of-the-art deep learning segmentation methods on three public breast ultrasound datasets show that our method has better performance on breast lesion segmentation. Furthermore, robustness analysis and external experiments demonstrate that our proposed AAU-net has better generalization performance on the segmentation of breast lesions. Moreover, the hybrid adaptive attention module can be flexibly applied to existing network frameworks.

Figures (8)

• Figure 1: Various BUS images and the segmentation results by U-net and our method. The red curve is the ground-truth boundary of the lesion. The yellow and green curves are the segmentation results of our method and U-net, respectively. It can be seen from these images that tumor morphology, blurred borders and similar surrounding tissue (background) severely affect the segmentation accuracy of breast lesions, especially for small and malignant tumors.
• Figure 2: Hybrid adaptive attention module (HAAM)
• Figure 3: The description of the adaptive attention U-net (AAU-net). The network is still a U-shaped network including four down-sampling and four up-sampling operations. Each stage consists of two hybrid adaptive attention modules (HAAM).
• Figure 4: Receptive fields captured by three convolutional layers. We can see that the receptive field obtained by$5 \times 5$ convolution layer is equivalent to 2 convolution layers with kernel size is $3 \times 3$, and the receptive field of dilated convolutions is the same as that of 5 convolution layers with kernel size is $3 \times 3$.
• Figure 5: P-R and ROC curves of different segmentation methods on BUSI and Dataset B.