Multimodal image registration by proxy image synthesis

描述(申请人提供)：图像配准是现代神经科学和临床医学中一项基本的重要能力。功能成像研究中的归一化，生长、衰老和疾病中形状变化的研究，术中图像上覆盖手术计划，以及几何失真校正都是图像配准的重要应用。在这两个主题中都有几十种注册需求 以及跨学科的应用。因此，图像配准性能的任何改进都将对科学界和临床界产生立竿见影的影响。尽管图像重建算法取得了许多进展，但使用多种模式(或组织对比度)来进行图像配准实际上是一个尚未开发的领域。最主要的框架是将对象的单个图像配准到目标的另一个单个图像，并且如果对象或目标的多个图像可用，则通过使用从单个图像配准导出的变换来配准它们。拟议的研究将开发、评估和验证一个非常简单的解释但相当激进的多模式注册的想法。 其基本思想是从具有与目标图像相同的组织对比度和亮度范围的被摄体图像合成一幅代理图像，然后使用诸如平方和差等常规度量来执行被摄体代理与目标之间的配准。初步结果表明，这种方法有显著的好处。在赠款中，我们将：1)通过探索其理论合理性以及整体方法的关键参数来优化“代理”多模式图像配准；2)将“代理”多模式图像配准应用于神经科学中的三个关键应用，以验证方法并制定最佳实践原则；以及3)编写开放源代码软件，使用“代理”图像概念执行图像合成、相似度计算、刚性和可变形配准。用于合成图像以用于用户最喜欢的图像配准方法的软件以及以优化方式执行整个“代理”配准过程的软件都将作为开放源码计算机代码公开提供。这项研究的结果将改变研究人员和从业者获取和使用神经科学研究和临床医学数据的方式，从而开创图像配准的新纪元。

DESCRIPTION (provided by applicant): Image registration is a fundamentally important capability in modern neuroscience and clinical medicine. Normalization in functional imaging studies, studies of shape changes in growth, aging, and disease, overlaying surgical plans on intraoperative images, and geometric distortion correction are examples of important applications of image registration. There are dozens of needs for registration in both intrasubject and intersubject applications as well. Therefore, any improvement in image registration performance will have an immediate impact on the scientific and clinical communities. Despite numerous advances in image reconstruction algorithms, the use of multiple modalities (or tissue contrasts) to carry out image registration is a virtually untapped area. The vastly dominant framework is to register a single image of the subject to another single image of the target, and if multiple images are available of either subject or target, they are registered by using the transformation derived from the single image registration. The proposed research will develop, evaluate, and validate a very simply explained but quite radical idea for multi-modal registration. The basic idea is to synthesize a "proxy" image from the subject image that has the same tissue contrast and intensity range as the target image and then use a conventional metric such as sum-of-square difference to carry out the registration between the subject proxy and target. Preliminary results demonstrate significant benefits in this approach. In the grant we will: 1) Optimize "proxy" multimodal image registration by exploring its theoretical justification as well a key parameters of the overall approach; 2) Apply "proxy" multimodal image registration to three key applications in neuroscience in order to validate the method and develop principles of best practice; and 3) Write open source software to carry out image synthesis, similarity computation, and rigid and deformable registration using the "proxy" image concept. Both the software to synthesize images for use in a user's favorite image registration method as well as software to carry out the entire "proxy" registration process in an optimized way will be made publicly available as open source computer code. The results of this research will lead to a new era in image registration by changing the way researchers and practitioners acquire and use data for neuroscientific studies and clinical medicine.