在68Ga EDTA动态PET分析中用正常区域取样和参数成像实现大脑肿瘤的统计分割研究

为了研究和评价用^68GaEDTA动态PET研究和评价脑肿瘤定量分析的可靠性和灵敏度，我们在本文中提出了估计容积分布（distribution volume:DV)和血脑屏障渗透率（k1)分布的线性参数成像模型。我们还用F统计学方法实现了把肿瘤从正常组织中分割出来的方法，用一个三参数双腔室模型描述用PET测量的数据，用于估计DV（K1/k2)和K1的主要计算公式为：Cpet=(K1 k2Vp)∫0^tCpds-k2∫0^tCpetds VpCp和∫0^tCpetds=(DV Vp)∫0^tCpds-(1/k2)Cpet (Vp/k2)Gp，这里的k2是脑内通过血脑屏障到脑外的渗透率，在参数成像中我们采用了一个可靠和如棒的基于像素的局域线性回归算法用于产生DV和K1图像。同样基于像素自由度为2和k-2的F统计学方法采用的计算公式为：F=（（（k-2)k/(2(k^2-1)))D^2。这里的D^2=(x-μ）′S^-1(x-μ），而μ和S分别表示脑肿瘤对侧正常区域内采集的样品的平均值和协方差，这些样品是按照二维空间{（K1，DV）}采集的，而常数k是取样的总数，在不同水平上阈值α，就可以得到不同置信度下的F统计图像，用这个方法对11个肿瘤病人进行了^68GaEDTA动态PET研究，研究结果表明：所有的DV，K1和F图像的质量都很好。而且用于产生DV，K1和不同置信度下的F图像的算法效率很高，容易实现，本研究方法中研究和发展的方法提供了一个有效的集成多维生理信息的工具，这个方法可以改善对肿瘤诊断和处理的灵敏度和特异性。

Statistical mapping of brain tumor using normal reference Regions and parametric images in 68Ga EDTA dynamic PET studies

To improve the reliability and sensitivity of quantitative analysis in the study and evaluation of brain tumor using 68Ga EDTA dynamic PET, a linear parametric imaging algorithm was developed in this study for estimation of both distribution volume (DV) and blood brain barrier permeability. F statistics was used for separating tumor from normal tissue. A two-compartmentad three-parameter model was used to describe the tracer kinetics measured by PET.The operational equations: Cpet=(K1+k2Vp) ∫t0 Cpds-k2 ∫t0 Cpet ds+VpCp and ∫t0 Cpet ds=(DV+Vp) ∫t0 Cpds-(1/k2)Cpet+(Vp/k2)Cp wereused to estimate K1 (permeability) and DV (=K1/k2), respectively. A reliable and robust linear regression algorithm with spatial constraint on parametric images was used to generate the K1 and DV images. Pixel-wise F statistics with 2 and k-2 degrees of freedom was calculated as: F= (((k-2)k/(2(k2-1)))D2with D2= (x-μ)'S-1(x-μ), where the sample is from two dimensional sample space {(K1, DV)} of reference regions in normal brain tissue, the sample size k is the number of pixels within the normal reference regions. μ and S are, respectively, the sample mean vector (K1, DV) and covarianee matrix. By setting critical a values at different levels. statistical significance level images were generated. The method was applied to eleven brain tumor 68Ga EDTA dynamic PET studies. Results showed that the DV, K1, and F images are of good image quality. The method tor generating K1. DV, F, and significance level images is of high computation efficiency and is easy to be implemented. The statistical model developed in the current study provided a tool to integrate the multi-dimensional physiological information. The normal reference region method and the integration of multi-physiological images may improve the sensitivity and specificity of brain tumor detection and evaluation of treatment.