近红外漫射光血流成像的建模与算法研究

Blood flow measurement in tissue microvascular network is crucial for detection and treatment of many diseases. Diffuse correlation spectroscopy/tomography (DCS/DCT) is emerging as a new technology for tissue blood flow measurement. At present, the DCT model is based on either the analytical solution or the finite element method derived from the partial differential equation. These two methods are unable to efficiently utilize the experimental data or the tissue information (e.g., geometry or internal components). As the results, the blood flow image is inaccurate and susceptible to the environmental noises. In this proposal, we plan to create a new DCS model and algorithm by combination of the medical images (MRI, CT or ultrasound) that reflects the geometrical and internal information. Specifically, the medical images are integrated with photon Monte Carlo simulation to obtain the tissue spatial characteristics, which are then input into the Nth-order algorithm (NL algorithm) that was recently created by us, subsequently forming the new DCT model. As for the numerical solution of DCT model (i.e., the solving the matrix), we plan to integrate the commonly-used image reconstruction methods with the physical feature of DCT model, so as to optimize the algorithm and obtain the accurate and stable images of blood flow. To overcome the limit of computing speed in iteration procedure for solving the matrix, we plan to accelerate the computing capacity with use of GPU parallel computing scheme, which will realize the dynamic blood flow tomography.

人体组织微血管的血流测量对于很多疾病的诊断和治疗至关重要，近红外漫射光相关谱和断层成像技术是测量组织血流的较新技术。目前DCT的模型主要是基于偏微分方程的解析法和有限元法，这两种方法不能充分地利用测量数据或被测组织的形态与内部成分等信息进行图像重建，因此血流成像质量低并且容易受到噪声的影响。本项目拟结合形态学的医学影像（磁共振、CT或超声影像等）来创建新的DCT血流成像模型和重建算法。具体方法是以最新提出的 “N阶线性算法”为基础，将形态学的医学影像和光子的蒙特卡罗仿真相结合来获得被测组织的空间特征；进一步，将空间特征和DCT数据融合来构建新型的DCT模型。在DCT模型的数值求解过程中，本项目拟将共性的图像重建方法与DCT的物理特征相结合来优化重建算法，由此获得精确和稳定的求解；针对迭代法速度较慢的特点，我们也使用GPU并行计算等方式进行重建算法的加速，从而实现动态的血流成像。

本项目主要研究两个科学问题，首先是如何创建新的DCT血流成像模型；另一个是开发DCT模型的重建算法。经过四年的努力，我们完成了研究计划中的内容：结合形态学的医学影像（磁共振）和“N阶线性算法”建立了新的DCT模型并通过光子的蒙特卡罗仿真获得空间特征；在此基础上，比较了多种图像重建算法并利用Bregman-TV方法获得血流成像；此外，我们使用GPU并行计算等方式进行了重建算法的加速。本项目开发的DCT模型和重建算法的精确性和速度通过计算机仿真和仿体实验得到的验证，并应用于脑血流成像等临床实践。本项目取得了较多研究成果，发表了15篇SCI检索论文，3篇EI检索论文和4篇中文论文，申请了3项中国发明专利，在国际和国内会议中做了11次报告，培养了9名博士和硕士研究生，技术成果也获得企业的资金支持，有重要医学应用价值。