Penalized discriminant analysis of [15O]-water PET brain images with prediction error selection of smoothness and regularization hyperparameters
Penalized discriminant analysis of [15O]-water PET brain images with prediction error selection of smoothness and regularization hyperparameters
复制标题
DOI：
10.1109/42.925291
复制
发表时间：
2001-05-01
影响因子：
10.6
通讯作者：
Strother, S
中科院分区：
文献类型：
--
作者：
Kustra, R;Strother, S
本文提出了一种灵活、全面的方法，利用惩罚型线性判别分析（PDA）对[O-15]-水正电子发射断层扫描（PET）脑图像进行分析，我们将其应用于20例受试者的扫描（8次扫描/受试者）进行手指移动任务，并分析：1）两个类以获得协方差归一化的基线激活图像，和2）8类平均受试者的时间结构，其中包含基线激活和时间依赖性的变化，在两个-我们通过将其扩展到五个具有不同平滑度的张量积B样条（TPS）基中来对所得到的图像施加空间平滑度，并进一步对噪声协方差矩阵进行脊型惩罚来正则化。PDA的歧视方法提供了一个概率框架内，预测误差（PE）估计得出。我们使用这些来优化TPS基和脊超参数（表示为等效自由度，EDF），我们使用现代回归工具获得了无偏，低方差PE估计（.632+ Bootstrap和交叉验证），并比较了1）TPS投影、平均归一化和未归一化扫描的PDA，以及2）有和没有额外预平滑的平均归一化扫描的PDA，通过检查PE和EDF之间的权衡，作为一个功能的基础选择和图像平滑，我们证明了PDA的效用，PE框架，奇异值分解和平滑TPS基地之间的关系在功能神经影像的分析。
We propose a flexible, comprehensive approach for analysis of [O-15]-water positron emission tomography (PET) brain images using a penalized version of linear discriminant analysis (PDA), We applied it to scans from 20 subjects (eight scans/subject) performing a finger movement task and analyzed: 1) two classes to obtain a covariance-normalized baseline-activation image, and 2) eight classes for the mean within subject temporal structure which contained baseline-activation and time-dependent changes in a two-dimensional canonical subspace, We imposed spatial smoothness on the resulting image(s) by expanding it in five tenser-product B-spline (TPS) bases of varying smoothness, and further regularized with a ridge-type penalty on the noise covariance matrix. The discrimination approach of PDA provides a probabilistic framework within which prediction error (PE) estimates are derived. We used these to optimize over TPS bases and a ridge hyperparameter (expressed as equivalent degrees of freedom, EDF), We obtained unbiased, low variance PE estimates using modern resampling tools (.632+ Bootstrap and cross validation), and compared PDA of 1) TPS-projected, mean-normalized and unnormalized scans and 2) mean-normalized scans with and without additional presmoothing, By examining the tradeoffs between PE and EDF, as a function of basis selection and image smoothing we demonstrate the utility of PDA, the PE framework, and the relationship between singular value decomposition and smooth TPS bases in the analysis of functional neuroimages.