Three-dimensional hybrid guidance system for cardiac interventional procedures

微创心脏手术是心血管疾病的常见治疗方法，涉及将柔性装置（例如导管或支架）插入心腔。x射线透视目前用于指导外科医生，因为该设备在x射线下高度可见，现代x射线系统提供实时（即无延迟）成像，大视场和出色的图像分辨率。然而，x射线图像提供的解剖信息很少，因为外科医生无法看到心腔和周围血管的位置，除非注射造影剂。此外，x射线图像是二维图像，因此图像内的物体可能相互重叠，这使得相对于复杂的心脏解剖结构，很难确定设备的准确位置。这导致手术时间延长，从而增加有害辐射剂量。为了添加这种解剖信息，混合制导系统已经开发出来，它将x射线信息与其他信息（例如来自计算机断层扫描）结合起来，在x射线图像的顶部添加阴影或轮廓。这些系统的缺点是它们仍然严重依赖x射线透视图像来提供指导，并且所有信息仍然是二维的。该项目的目的是开发一种新的3D混合制导系统，优于这些现有的方法。它将为外科医生提供3D信息，提高他们的效率，从而减少x射线暴露。它还将使用额外的3D引导设备，如电解剖测绘（EAM）系统，以减少x射线图像的频率，从而进一步减少x射线暴露。EAM系统使用弱磁场而不是有害的x射线辐射，因此可以在整个过程中打开。EAM系统的主要用途是绘制心脏内的电生理活动。但它也可以跟踪心脏腔内的导管，并创建心脏腔的低分辨率3D模型。当使用EAM系统时，不可能清晰地可视化3D血管结构，并且一些设备（如支架和气球）可能无法跟踪。因此，需要提出具有x射线信息的混合系统。为了开发这个系统，我们将使用先进的计算机视觉技术来检测设备，并从x射线图像中提取3D血管模型，然后将这些模型与EAM系统内现有的3D模型融合，以提供完整的信息来指导手术过程。由于低剂量x射线图像中存在高水平的噪声和重叠物体的干扰，实现准确和鲁棒的实时检测是一项具有挑战性的任务。为了应对这一挑战，提出了一种利用图像分类器同时检测电极导管和导线上的器件的新方法。因为所有的设备都是附着在电线上的物体，我们的基于学习的图像分类器只需要搜索沿着电线状物体的区域。此外，我们的方法还将能够解决在不同坐标系下测量的两个系统中的3D模型之间精确对齐的挑战。定位是基于在EAM系统和x射线系统中跟踪同一设备的3D位置。由于可以使用EAM系统作为主要制导工具，并且使用较少的x射线图像，因此我们提出的系统将大大减少x射线辐射暴露。这将使病人受益，因为x射线辐射可能在他们以后的生活中导致癌症。我们将与雅培医疗英国有限公司合作，目标是利用雅培的EAM系统开发和调整我们的方法，以便在不久的将来制作研究原型。但我们的理论贡献并不局限于EAM系统，而且可以用于混合x射线图像与其他图像制导系统，如3D回波成像，以及未来的机器人手术系统。

Minimally invasive cardiac surgeries are the common treatment for cardiovascular disease, involving the insertion of flexible devices (e.g. catheters or stents) into heart chambers. X-ray fluoroscopy is currently used to guide surgeons as the devices are highly visible under X-rays and modern X-ray systems provide real-time (i.e. with no lag) imaging, a large field-of-view and excellent image resolution. However, X-ray images offer very little anatomical information as surgeons cannot see where the heart chamber is and its surrounding blood vessels, unless contrast agents are injected. Furthermore, X-ray images are 2D images and so objects inside the image could overlap each other making it difficult to determine the accurate position of devices relative to the complex heart anatomy. This results in extended procedure times and thus additional harmful radiation doses. To add this anatomical information, hybrid guidance systems have been developed which combine the X-ray information with other information (e.g. from computerised tomography) to add the shadows or contours on the top of the X-ray images. The drawbacks of these systems are that they still heavily rely on X-ray fluoroscopic images to provide guidance, and all information is still 2D.The aim of this project is to develop a new 3D hybrid guidance system superior to these existing approaches. It will provide 3D information to surgeons, increasing their efficiency and thus reducing X-ray exposure. It will also use additional 3D guidance equipment such as the electroanatomical mapping (EAM) system to reduce the frequency of X-ray images, and so further reduce X-ray exposure. The EAM system uses a weak magnetic field rather than harmful X-ray radiation and so it can be switched on throughout the procedure. The primary use of the EAM system is to map electrophysiological activities within the heart. But it also can track catheters within a heart chamber and create low-resolution 3D models of heart chambers. It is not possible to visualise the 3D blood vessel structures clearly when using the EAM system and also some of devices such as stents and balloons might not be tracked. Hence the need for the proposed hybrid system with X-ray information.To develop this system we will use advanced computer vision techniques to detect devices and extract 3D blood vessel models from X-ray images, and then fuse these with existing 3D models inside the EAM system to provide the completed information to guide the procedure. Due to the high-level of noise present in low-dose X-ray images and the interference from overlapping objects, it is a challenging task to achieve accurate and robust detection in real-time. To meet the challenges, a novel approach is proposed to simultaneously detect the electrode catheters by the electrode pattern and the device on the wire by an image classifier. Since all devices are objects attached to the wires, our learning-base image classifiers will only need to search the areas along the wire-like objects. Furthermore, our approach will also be able to solve the challenge of the accurate alignment between 3D models in two systems measured in different coordinate systems. The alignment is based on tracking the 3D position of the same device in both an EAM system and an X-ray system. As it is possible to use the EAM system as the main guidance tool and use less frequent X-ray images, our proposed system will significantly reduce X-ray radiation exposure. This will benefit patients as X-ray radiation might cause the cancer in their later life. We will partner with Abbott Medical UK Ltd, and aim to develop and adapt our approach using Abbott's EAM system so that a research prototype can be made in the near future. But our theoretical contributions will not limited to the EAM system, and could be used to hybridise X-ray images with other image-guidance systems, such as the 3D echo imaging, as well as future robotic surgery systems.