Vessel-based Framework for Joint Image Registration and Segmentation

可变形图像配准和分割可以为医生提供更好的图像质量和重要信息，因此成为医疗规划和诊断的重要组成部分。在该方案中，我们将研究新的可变形图像处理技术，包括智能关节图像配准和分割、血管分割和非均匀软组织的可变形图像配准。本研究的成果将为减少医生的阅读时间，提高诊疗计划的准确性和质量提供有效的途径。开发的技术将用于肝脏外科手术中的治疗规划和指导，也可以扩展到肺、大脑和心脏。 大多数基于物理的可变形配准研究工作都假定组织是均匀的。然而，这一假设在许多医学应用中并不成立。例如，为了帮助肿瘤医生/外科医生确定恶性肿瘤的状态，我们经常需要将区域与嵌入的肿瘤对齐，在这些情况下不能假设同质性，因为肿瘤和周围组织具有非常不同的僵硬和弹性属性。我们在肝脏图像配准的研究工作中观察到，假设同质性会导致配准精度有很大的差异。为了解决这一问题，我们提出将整个感兴趣区域划分为更小的子区域，并在配准目标函数中动态调整每个子区域中血管段和分叉点的权重，从而能够实现包括肿瘤邻域在内的整个区域的准确配准。 在基于特征的图像配准技术中，准确的图像配准在很大程度上依赖于准确的图像分割。血管是许多器官的重要结构，可以为图像配准提供基础信息。它们也是在器官内部定位治疗靶点的很好参考。准确的血管分割不仅可以为配准提供有用的信息，而且可以用于检测有血管区域的配准错误。由于可变形软组织/器官的复杂性和多变性，准确快速的分割仍然是一项非常具有挑战性的任务。血管分割有一些独特的困难，包括血管缺失、血管断开、血管接触、噪声、边界模糊和不完整、图像对比度不足等。在这份提案中，我们将研究这些挑战，并使用由预先获取的高质量图像产生的特定于患者的指导模型来调查问题。

Deformable image registration and segmentation can provide the doctors with better image quality and vital information and therefore is becoming the essential part in medical treatment planning and diagnosis. In this proposal we will investigate new deformable image processing techniques including intelligent joint image registration and segmentation, vessel segmentation and deformable image registration of inhomogeneous soft tissues. The outcome of this research will provide an effective way to reduce doctors' reading time and improve accuracy and quality of diagnosis and treatment planning. The technologies developed will be used for the treatment planning and guidance in surgical procedures on the liver and can also be extended to the lungs, the brain and the heart. Most of physics based deformable registration research works assume that the tissues are homogeneous. However, this assumption is not valid in many medical applications. For example, in order to help oncologist/surgeons to determine the status of malicious tumors, we often need to align the regions with embedded tumors, homogeneity cannot be assumed in these cases because the tumor and the surrounding tissues have very different properties of stiffness and elasticity. Assumption of homogeneity would result in large discrepancy in registration accuracy, as we have observed in our research work on image registration of liver images. To address this problem, we propose to partition the whole region of interest into smaller sub-regions and dynamically adjust weights of vessel segments and bifurcation points in each sub-region in the registration objective function so that we are able to achieve accurate registration for the whole region including the neighborhood of the tumors. In feature based image registration techniques, accurate image registration to a large extent relies on accurate image segmentation. Blood vessels are critical structures of many organs and can provide fundamental information for image registration. They are also good references for localizing treatment targets deep inside organs. Accurate segmentation of vessels not only can provide useful information for registration, but also can be useful for detecting registration errors in regions with vessels. Accurate and fast segmentation is still a very challenging task due to the complexity and variability of deformable soft tissues/organs. There are some unique difficulties that are associated with vessel segmentation including missing vessels, disconnected vessels, touching vessels, noise, vague and incomplete boundaries, inadequate contrast of images. In this proposal, we will examine these challenges and investigate the problems using patient-specific guidance models produced from pre-acquired high quality images.