Multigrid Optical Diffusion Tomography

最近，人们对基于光的医学成像模式的发展越来越感兴趣，而不是基于超声、x射线计算机断层扫描（CT）或磁共振（MRI）的更传统的模式。基于光的成像具有无创、安全、只需要廉价仪器的优点。传统观点认为，光不能穿过组织，但事实上，近红外光确实可以穿过组织，因此，原则上，可以用来形成组织内部的图像。然而，通过组织的光是高度散射的，因此使用传统方法无法计算图像。这项研究涉及新的计算算法的发展，该算法可用于从散射光的测量中形成三维图像。这些方法统称为光学扩散层析成像，具有提供安全、廉价和便携式成像仪器的潜力。光学扩散成像的实际实现要求以计算易于处理的方式解决复杂的非线性逆问题，以获得准确的图像。在这项研究中，开发了新的多网格优化方法，这些方法有可能大大加快重建速度。这些新方法应用于贝叶斯框架，并使用新技术来模拟含有空洞，不透明组织或荧光显像剂的区域。此外，光学扩散层析成像是一类更广泛的重要逆问题的代表；因此，本研究开发的方法在图像配准、运动估计、环境感知和无损评价等问题中具有广泛的应用前景。

Recently, there has been growing interest in the development of medical imaging modalities based on light, as opposed to more conventional modalities based on ultrasound, X-ray computed tomography (CT), or magnetic resonance (MRI). Imaging based on light has the advantages of being non-invasive, safe, and requiring only inexpensive instrumentation. Conventional wisdom holds that light can not pass through tissue, but in fact, near infra-red light does pass through tissue and can therefore, in principle, be used to form images of the interior of the tissue. However, light passing through tissue is highly scattered, so that an image can not be calculated using conventional methods. This research is concerned with the development of new computational algorithms that can be used to form three-dimensional images from measurements of scattered light. These methods, known collectively as optical diffusion tomography, have the potential to provide safe, inexpensive, and portable imaging instruments.Practical realization of optical diffusion imaging requires that a difficult non-linear inverse problem be solved in a computationally tractable manner to yield accurate images. In this research, new multigrid optimization methods are developed which have the potential to dramatically speed reconstruction. These new methods are applied in a Bayesian framework and use novel techniques to model the regions containing voids, opaque tissue, or fluorescence imaging agents. Moreover, optical diffusion tomography is representative of a broader class of important inverse problems; so that the methods developed in this research have application in problems ranging from image registration and motion estimation to environmental sensing and non-destructive evaluation.