Wang Feng, Guan Bo, Zhao Jianchang
Institute of Medical Robotics and Intelligent Systems, Tianjin University, Tianjin, 300392, China.
Key Laboratory for Mechanism Theory and Equipment Design of Ministry of Education, Tianjin University, Tianjin, 300354, China.
J Robot Surg. 2025 Aug 10;19(1):472. doi: 10.1007/s11701-025-02644-3.
Vascular contrast enhancement is crucial for early disease diagnosis and surgical precision in robotic surgery imaging. Traditional white-light imaging often fails to distinguish blood vessels due to the spectral similarity between vessels and surrounding tissues. Although techniques like narrow-band imaging improve contrast, they require specialized hardware and exhibit inconsistent performance across different surgical environments. To address these limitations, we propose StarVasc, a novel lightweight framework for unsupervised vascular contrast enhancement tailored for robotic surgical vision systems. StarVasc leverages an unpaired learning strategy based on a compact generative adversarial network. The generator incorporates a star operation module, enabling hyper-dimensional feature expansion. This operation implicitly maps input images into an exponentially high-dimensional nonlinear feature space, facilitating efficient representation of fine-grained vascular structures without increasing model complexity. In addition, we design a Spectral Feature Enhancement Module (SFEM) to further refine vascular detail. Acting as a narrow-band feature extractor, SFEM implicitly learns spectral cues without requiring hyperspectral input. It operates in a self-supervised reconstruction paradigm, ensuring that the extracted features are semantically aligned with vascular structures. Integrated within an encoder-decoder architecture, SFEM enhances vessel clarity and edge continuity in the output images. Extensive experiments demonstrate that StarVasc consistently outperforms both traditional enhancement techniques and recent deep learning methods across no-reference quality metrics and visual evaluations. Without relying on specialized hardware, StarVasc provides an adaptive, clinically viable solution for real-time vascular enhancement in robotic surgical imaging, contributing to improved visual perception and surgical safety in automated or robot-assisted interventions.
血管造影增强对于机器人手术成像中的早期疾病诊断和手术精度至关重要。传统的白光成像由于血管与周围组织之间的光谱相似性,常常无法区分血管。尽管窄带成像等技术提高了对比度,但它们需要专门的硬件,并且在不同的手术环境中表现出不一致的性能。为了解决这些局限性,我们提出了StarVasc,这是一种新颖的轻量级框架,用于为机器人手术视觉系统量身定制的无监督血管造影增强。StarVasc利用基于紧凑生成对抗网络的非配对学习策略。生成器包含一个星形操作模块,能够进行超维特征扩展。此操作将输入图像隐式映射到指数级高维非线性特征空间,有助于在不增加模型复杂性的情况下有效表示细粒度血管结构。此外,我们设计了一个光谱特征增强模块(SFEM)来进一步细化血管细节。作为一个窄带特征提取器,SFEM隐式学习光谱线索,而无需高光谱输入。它在自监督重建范式中运行,确保提取的特征在语义上与血管结构对齐。集成在编码器-解码器架构中,SFEM增强了输出图像中血管的清晰度和边缘连续性。大量实验表明,在无参考质量指标和视觉评估方面,StarVasc始终优于传统增强技术和最近的深度学习方法。无需依赖专门的硬件,StarVasc为机器人手术成像中的实时血管增强提供了一种自适应、临床上可行的解决方案,有助于在自动化或机器人辅助干预中改善视觉感知和手术安全性。