PARAMETRIC RECONSTRUCTION IN DIFFUSE OPTICAL IMAGING

这个子项目是许多研究子项目中利用 资源由NIH/NCRR资助的中心拨款提供。子项目和 调查员(PI)可能从NIH的另一个来源获得了主要资金， 并因此可以在其他清晰的条目中表示。列出的机构是 该中心不一定是调查人员的机构。 用于光学成像的常规层析成像方法最适合于包含来自大量源-探测器对但波长很少(即，空间上丰富，但光谱上稀疏)的测量的数据集。由于技术的复杂性，定量宽带光谱技术很难在传统的成像配置中实现。因此，这些传统的层析成像方法不容易使用仅使用有限数量的源-探测器对的宽带光谱学技术来实现。我们在稀疏层析成像(Pr)参数重建方面所做的努力，是针对那些在光谱上“丰富”但在空间上“稀疏”的数据集。 有证据表明，相对简单的断层成像方法可能是对手持宽带光谱的需求和癌症应用中对深度切片成像的需求之间的良好折衷。基于我们使用正向有限元建模程序对乳腺临床数据的分析，我们认为乳腺病变的光学属性的空间范围远远大于用传统放射学方法测量的肿瘤的结构范围(图XXX)。左上角的面板代表了肿瘤的传统观点：光学特性的扰动被巧妙地分割为背景(下标为BKG)和异质性(下标为HET)。在典型的漫反射光谱测量中，源(S)和探测器(D)被串联扫描整个组织。提供了实际漫反射光谱测量的结果(圆圈)，并与使用有限元方法提供的SDA辐射传输模型的正向模型模拟进行了比较(平方)。我们发现，没有一组物理上的模拟光学特性可以复制临床测量。在图XXX的右侧，我们展示了肿瘤的不同视图：光学性质的梯度。通过改变光学特性分布的属性，临床数据和模拟数据共享更接近的一致性(右下角)。在对10名患者的初步研究中，我们进一步发现，当光学特性分布的空间范围远远大于放射学大小估计时，建模结果与临床数据之间的一致性最好。对于恶性病变尤其如此。病变的较大空间范围放松了可视化目标所需的源-探测器对的数量。此外，通过将光学特性的空间分布限制为由高斯(或其他)分布给出的空间分布，也减少了重建变量的数量。使用这种方法的简单重建也消除了对正则化参数的需要，这种方法我们称之为“稀疏层析成像”。 我们建议的稀疏断层扫描程序可以用于任何成像数据，尽管它可能在成像能力有限的系统中工作得最好。我们将首先使用我们的DOS/I仪器对乳腺病变进行扫描。数据将在VTS内进行处理，以恢复为半无限均匀扩散模型提供最小二乘拟合的光学性质。接下来，我们调用基于扩散的有限元模型，该模型具有光学吸收和散射特性相对于背景光学特性的扰动的高斯空间分布。然后用该模型用有限元方法生成模拟数据。接下来，对模拟数据进行相同的最小二乘处理，使其适用于相同的半无限均匀扩散模型。最后，我们将在卡方意义上将我们的临床观察与我们的模拟结果进行比较。这个过程将反复进行，直到卡方最小化。我们意识到这一过程实际上是在“模糊”目标的位置。然而，我们不认为这种对肿瘤的描述是非物理的，因为众所周知，与放射学确定的大小相比，肿瘤的边缘可能相当广泛。

This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. Conventional tomographic approaches for optical imaging are best suited for data sets that contain measurements from a large number of source-detector pairs but few wavelengths (i.e., spatially 'rich,' but spectrally 'sparse'). Due to technical complexities, quantitative broadband spectroscopic technologies are difficult to implement in traditional imaging configurations. As a result, these conventional tomography approaches are not easily implemented with broadband spectroscopy techniques that use only a limited number of source-detector pairs. Our effort in 'sparse tomography,' pr 'parametric reconstruction,' is geared towards data sets that are spectrally 'rich' but spatially 'sparse.' There is evidence that a relatively simple tomographic approach may be a good compromise between the demand for handheld-acquired broadband spectra and the need for depth-sectioned imaging in cancer applications. Based on our analysis of breast clinical data with a forward finite element modeling procedure, we believe the spatial extent of a breast lesion's optical properties is much larger than the structural confines of the tumor as measured by conventional radiological methods (Figure XXX). The top-left panel represents the traditional view of a tumor: a perturbation in optical properties neatly segmented between background (subscript 'bkg') and heterogeneity (subscript 'het'). In a typical Diffuse Optical Spectroscopy measurement, the source (S) and detector (D) are scanned in tandem across a tissue. The results of an actual Diffuse Optical spectroscopy measurement are provided (circles) in comparison with a forward model simulation using a radiative transport model of the SDA provided by the Finite Element Method (squares). We discovered that there was no physical set of simulated optical properties that could replicate the clinical measurement. On the right side of Figure XXX, we present a different view of the tumor: a gradient in optical properties. By varying the properties of the optical property distribution, the clinical and simulated data share a much closer agreement (bottom-right). In a pilot study of 10 patients, we further found that the best agreement between the modeling results and the clinical data is achieved when the spatial extent of the optical property distribution is much larger than radiological size estimates. This is especially true for malignant lesions. The larger spatial extent of the lesion relaxes the number of source-detector pairs necessary to visualize the target. Moreover, by constraining the spatial distribution of optical properties to those given by a Gaussian (or other) distribution also reduces the number of reconstructed variables. Simple reconstructions using such an approach, which we refer to as "sparse tomography," also eliminates the need for regularization parameters. Our proposed sparse tomography procedure can be used with any imaging data, although it may work best with systems that have limited imaging capabilities. We will first use our DOS/I instrument to take scans of breast lesions. The data will be processed within the VTS to recover the optical properties that provide a least squares fits to semi-infinite homogenous diffusion models. Next we invoke a Finite Element Model diffusion-based model of a breast lesion with a Gaussian spatial distribution of a perturbation in optical absorption and scattering properties relative to the background optical properties. Simulated data will then be generated with this model using finite element methods. Next the simulated data will be processed with the same least-squares fit to the same semi-infinite homogenous diffusion model. Finally, we will compare our clinical observation with our simulated result, in a chi-squared sense. This process will be iterated until the chi-squared is minimized. We realize that this procedure is in effect "blurring" the location of the target. However we do not believe that this description of a tumor is unphysical because it is well known that the margins of tumors can be quite extensive compared to the radiologically-determined size.