DBF-Net: A Dual-Branch Network with Feature Fusion for Ultrasound Image Segmentation

TL;DR

UBBS-Net, a dual-branch deep neural network that learns the relationship between body and boundary for improved segmentation, is introduced and a feature fusion module to integrate body and boundary information is proposed.

Abstract

Accurately segmenting lesions in ultrasound images is challenging due to the difficulty in distinguishing boundaries between lesions and surrounding tissues. While deep learning has improved segmentation accuracy, there is limited focus on boundary quality and its relationship with body structures. To address this, we introduce UBBS-Net, a dual-branch deep neural network that learns the relationship between body and boundary for improved segmentation. We also propose a feature fusion module to integrate body and boundary information. Evaluated on three public datasets, UBBS-Net outperforms existing methods, achieving Dice Similarity Coefficients of 81.05% for breast cancer, 76.41% for brachial plexus nerves, and 87.75% for infantile hemangioma segmentation. Our results demonstrate the effectiveness of UBBS-Net for ultrasound image segmentation. The code is available at https://github.com/apple1986/DBF-Net.

Figures (8)

• Figure 1: Three different convolution neural network architectures: (a) the classic encoder-decoder structure, (b) a network structure with added boundary information, and (c) the proposed structure in this paper, which do feature interaction and enhancement between body decoder and boundary decoder. The red curve delineates the boundaries of the lesions.
• Figure 2: The proposed DBF-Net. 8x and 2x are upsampling ratios. C and + means concatenation and addition, respectively. Our model adopts the traditional encoder-decoder architecture. After the encoder, we utilize an ASP_OC module to extract multi-scale features. The feature fusion and supervision block (FFS) focuses on explicitly modeling body and boundary at the feature level through a set of convolution operations, feature fusion block, and supervision.
• Figure 3: Details of boundary and body feature fusion block. Conv means successive convolution operations.
• Figure 4: Comparison of qualitative results between U-Net, DeepLabV3+, LinkNet, UNeXt, DBBS-Net and the proposed method for breast cancer segmentation using BUSI dataset, brachial plexus nerves segmentation using UNS, infantile hemangioma segmentation using UHES. The red curve outlines the boundaries of the lesions. The first and second rows show results from the BUSI dataset, while the third and fourth rows depict results from UNS, and the fifth and sixth rows present results from UHES.
• Figure 5: P-R and ROC curves of DBF-Net, LinkNet, U-Net, DeepLabV3+, UNeXt, and DBBS-Net on BUSI, UNS, and UHES.