A Novel Diagnostic Tool: from Structural Health Monitoring to Tissue Quality Prediction

随着生活质量的不断提高，平均寿命将继续增加。坏消息是，在衰老的身体中，由于磨损造成的组织退化是不可避免的，而且因人而异。幸运的是，最近科学技术的进步使我们能够朝着个性化医疗的方向努力。这种方法需要来自广泛专业学科的研究人员的集体努力。该项目由来自英国四所不同大学的跨学科团队组成，他们拥有不同的专业领域，旨在预测患者特定的组织质量，这对于制定治疗计划至关重要。虽然我们在这项研究中主要关注的是骨组织，但开发的框架将适用于其他具有多孔或复杂微观结构的组织。为了实现这个项目的目标，我们必须克服一些挑战。首先，我们需要更好的数学模型来准确、自动地提取组织微观结构，也需要可靠的数学方法来比较正常图像（比如1024 x 1024像素）或包含形状和嵌入复杂结构的3D图像中的两个不同样本。尽管人们期望现有的软件可以完成这些任务，但现实情况是，这些任务比我们想象的要困难，因为来自真实扫描的图像包含噪声。为了对抗噪声，我们使用了一种新的数学技术，称为变分方法，它实际上构建了一个涉及整个图像的能量，并将其最小化。在这样做时，获得无噪声重建，并确定底层组织的精确位置。该项目将开发高效和稳健的模型，仅提取局部特征。类似地，我们可以在所谓的共配准问题中构造不同的能量来比较两个图像。我们的新想法是使用更多的数学方法和更少的统计估计方法来达到更高的准确性和鲁棒性。位于利物浦的CMIT研究中心专门研究针对不同问题情境的一系列数学模型。其次，一旦成像提取微观结构几何，我们来自爱丁堡和赫瑞瓦特大学的团队将开发并使用计算力学方法来评估组织的材料特性。由于组织的质量取决于其微观结构的样子，因此回答诸如：它是否多孔，不同的固体部分是否连接不良以及大多数孔隙是否在同一方向排列等问题，应该表明组织的强度。我们将检查这些微观结构，然后在计算机上进行一系列力学测试，以获得告诉我们组织何时以及如何受损或失效的特性。然后，我们研究了微观结构几何与力学行为的关系。我们将使用一种称为均质化的过程，该过程将能够从微观水平的知识预测宏观（组织）水平的特性。均质化预测组织衰竭之前还没有尝试过，这是一个重大的挑战。最后，预测的性质必须通过成像和定制的体外实验相结合来验证。验证过程将在达勒姆完成，将评估我们模型的准确性并提供模型改进。一旦验证，我们的方法可以放心地用作直接从图像评估组织特性的工具。因此，提议的项目包含技术和先进的数学组件以及在非侵入性成像适用的场景中的实际应用。随着成像技术变得越来越好，越来越便宜，越来越广泛，这项工作的新应用前景是巨大的。

As quality of life constantly improves, the average lifespan will continue to increase. The bad news is that tissue degradation due to wear and tear in an aged body is inevitable and is different from person to person. Fortunately recent advances in sciences and technology have enabled us to work towards personalised medicine this approach calls for a collective effort of researchers from a vast spectrum of specialised subjects. This project, by an interdisciplinary team from four different UK Universities with distinct areas of expertise, aims to predict patient-specific tissue quality which is essential in devising treatments plans. While our primary concern in this study is the bone tissue, the developed framework will apply to other tissues having porous or complex microstructure.To achieve the aim of the project, we have to overcome a few challenges. Firstly we need better mathematical models to extract the tissue microstructure accurately and automatically and also reliable mathematical methods for comparing two different samples from either a normal image (say with 1024 x 1024 pixels) or a 3D image containing shapes and embedded complex structures. Although one would expect that existing softwares can do these tasks, the reality is that these tasks are harder than we think as the images from a realistic scan contains noise. To counter noise, we use a novel mathematical technique called the variational method that actually constructs an energy quantity involving the whole image and minimises it. In doing so, a noise-free reconstruction is obtained and the precise location of the underlying tissue is identified. This project will develop efficient and robust models to extract local features only. Similarly we can construct different energies to compare two images in the so-called co-registration problem. Our new idea is to use more mathematical approaches and less statistical estimation ideas to achieve more accuracy and robustness. The CMIT research centre at Liverpool specialises in a range of mathematical models for different problem scenarios. Secondly once imaging extracts microstructural geometry, our team from Edinburgh and Heriot-Watt will develop and use a computational mechanics approach to evaluate the tissues' material properties. Since tissue quality depends on what its microstructure looks like, answers to questions such as: is it very porous, are different solid parts poorly connected and are most of the pores aligned in the same direction, should indicate how strong the tissue is. We will examine these microstructures and then conduct range of mechanical tests on the computer so as to obtain properties that tell us when and how the tissue will get damaged or fail. We then study how the microstructural geometry relates to the mechanical behaviour. We will use a process called homogenisation that will enable prediction of properties at macro (tissue) level from knowledge of micro level. Homogenisation to predict tissue failure has not been attempted before and presents a major challenge. Finally, the predicted properties must be validated from a combination of imaging and bespoke in-vitro experimentation. The validation process, to be done in Durham, will evaluate the accuracy of our models and provide model refinements. Once validated, our methodology can be used with confidence as a tool to evaluate tissue properties straight from images.Thus the proposed project contains both technical and advanced mathematical components and real applications in scenarios where non-invasive imaging is applicable. With imaging technology getting better, cheaper and more wide-spread, the prospects for novel applications of this work are immense.