用于高质量三维重建的偏振计量双目内窥成像技术的研究

Minimally-invasive surgery (MIS) aims to reduce tissue trauma by using small incisions to access the body with surgical endoscopes to see the interior. However, lack of depth information in typical endoscopic imaging systems mean it is difficult to perceive 3D shape of tissue. State-of-the-art binocular stereo endoscopy relies on disparities from image texture feature detection and matching to recover 3D shape, and is therefore ill-suited for tissue areas where featureless textures, occlusions from surgical instruments or view dependant appearance present. .Polarimetry refers to the analysis of the polarization, or oscillation state, of a light wave. Unpolarized light (oscillation states of light waves in the beam have no correlation) becomes partially polarised in reflection. 3D orientation of the reflection surface determines polarization state of the reflected light. Therefore imaging polarimetry can be used to compute orientation of the surface at every pixel. Latest progress has shown hybridizing polarimetry with multi-view stereo technique can recover the object’s shape with high-quality and high-resolution. The technique does not rely on image features, and therefore well-suited for surgical use, where low-contrast and homogeneous features of tissue limit the use of conventional computer vision methods. However, the challenges are lack of endoscope imaging hardware and a computational method suitable for biological tissues and specific endoscope illumination-imaging geometry..The aim of the proposed project is to develop a new polarimetry-enhanced endoscopic system for high resolution and artefact-free 3D tissue shape recovery. The outcome of this project will be a new polarimetry-enhanced binocular endoscopic system for high resolution and artefact-free 3D tissue shape recovery, which could result in advances in the field of image guided minimally invasive surgery and robotic surgery.

微创手术中外科医生需要利用内窥镜对体内进行观察。常规内窥镜通常难以获取组织表面形状等关键三维信息。高质量（即高分辨、少瑕疵）数字化的组织表面三维信息一方面可以为部分病变的诊断提供客观定量的依据，另一方面也是新一代图像引导微创手术与机器人微创手术等领域发展的重要基础。而现有的三维内窥镜技术依然有一定局限性，重建质量仍需改善。组织的偏振性质含有其表面法线方向的信息。本课题提出一种可用于高质量三维重建的新型偏振计量双目内窥成像技术。这项技术可以融合由偏振计量成像得到的稠密法线方向信息与由多视角三维重建得到的稀疏深度信息，实现稠密三维重建。该技术可以适用于无显著纹理特征、镜面反射过强、以及部分被器械遮挡的组织区域的深度信息恢复，因此可以显著提高三维重建质量，为改善当前微创手术实践、促进未来图像引导微创手术与智能化机器人手术等重要技术的发展奠定基础。

内窥镜是微创手术的核心设备之一。发展能够提供高质量图像信息的内窥成像技术，是改进当前的微创手术实践、促进未来新一代微创手术技术发展的迫切需求。本项目重点探索了利用偏振测量技术提升内窥成像质量，研究了商用单目和双目内窥镜偏振性质，揭示了其退偏性质的本质，据此提出了兼容蓝宝石窗的保偏内窥镜设计。研究了快照式斯托克斯矢量偏振测量成像技术，实现了视频级帧率的全斯托克斯矢量偏振测量，使针对运动非刚体目标偏振测量成像奠定了基础。利用保偏内窥镜完成了实时术中偏振测量内窥成像系统，成功开展了偏振测量内窥成像的首个人体实验，显示了偏振测量信息可增强病灶对比度最高至1个数量级。揭示了不同类型组织相位延迟成分的空间分布不同，是肿瘤组织与正常组织相位延迟形成对比度的根源。完成了可用于双目偏振测量内窥成像的实验系统，确定了该系统的标定、偏振测量图像校准、降噪去马赛克预处理方法，验证了利用偏振信息完成组织表面三维形状重建算法。提出了基于偏振测量成像的手术烟雾去除方法，表明了基于偏振测量成像技术去除手术烟雾、提升内窥图像质量的可行性，提出并验证了基于对抗深度卷积神经网络模型的由单一波段下偏振测量图像生成色彩图像的方法，为进一步使用偏振测量技术辅助高质量内窥成像及三维重建奠定了基础。