RESUMO
Minimally invasive surgery, which relies on surgical robots and microscopes, demands precise image segmentation to ensure safe and efficient procedures. Nevertheless, achieving accurate segmentation of surgical instruments remains challenging due to the complexity of the surgical environment. To tackle this issue, this paper introduces a novel multiscale dual-encoding segmentation network, termed MSDE-Net, designed to automatically and precisely segment surgical instruments. The proposed MSDE-Net leverages a dual-branch encoder comprising a convolutional neural network (CNN) branch and a transformer branch to effectively extract both local and global features. Moreover, an attention fusion block (AFB) is introduced to ensure effective information complementarity between the dual-branch encoding paths. Additionally, a multilayer context fusion block (MCF) is proposed to enhance the network's capacity to simultaneously extract global and local features. Finally, to extend the scope of global feature information under larger receptive fields, a multi-receptive field fusion (MRF) block is incorporated. Through comprehensive experimental evaluations on two publicly available datasets for surgical instrument segmentation, the proposed MSDE-Net demonstrates superior performance compared to existing methods.
RESUMO
Accurate surgical instrument segmentation can provide the precise location and pose information to the surgeons, assisting the surgeon to accurately judge the follow-up operation during the robot-assisted surgery procedures. Due to strong context extraction ability, there have been significant advances in research of automatic surgical instrument segmentation, especially U-Net and its variant networks. However, there are still some problems to affect segmentation accuracy, like insufficient processing of local features, class imbalance issue, etc. To deal with these problems, with the typical encoder-decoder structure, an effective surgical instrument segmentation network is proposed for providing an end-to-end detection scheme. Specifically, aimed at the problem of insufficient processing of local features, the residual path is introduced for the full feature extraction to strengthen the backward propagation of low-level features. Further, to achieve feature enhancement of local feature maps, a non-local attention block is introduced to insert into the bottleneck layer to acquire global contexts. Besides, to highlight the pixel areas of the surgical instruments, a dual-attention module (DAM) is introduced to make full use of the high-level features extracted from decoder unit and the low-level features delivered by the encoder unit to acquire the attention features and suppress the irrelevant features. To prove the effectiveness and superiority of the proposed segmentation model, experiments are conducted on two public surgical instrument segmentation data sets, including Kvasir-instrument set and Endovis2017 set, which could acquire a 95.77% Dice score and 92.13% mIOU value on Kvasir-instrument set, and simultaneously reach 95.60% Dice score and 92.74% mIOU value on Endovis2017 set respectively. Experimental results show that the proposed segmentation model realizes a superior performance on surgical instruments in comparison to other advanced models, which could provide a good reference for further development of intelligent surgical robots. The source code is provided at https://github.com/lyangucas92/Surg_Net.