THE SPECTRAL SIGNATURE METHOD FOR BRAIN PATTERN ANALYSIS

本研究项目的总体目标是发展和 验证一种基于计算机的方法，以帮助分析 产生的三维人脑图像中的功能模式 由PET或SPECT摄像机拍摄。这种方法与健康问题有关，因为它 (1)描述精神障碍患者的脑功能障碍模式 疾病；(2)涉及药物引起的代谢活动的模式， 药物滥用或其他刺激对大脑解剖区域的影响；(3)检测 两组之间的模式差异；(4)检测 异质组在临床分类中的作用 大脑。到目前为止，大多数方法都依赖于 解剖感兴趣区(ROI)以减少功能性 要分析的信息，但在这个过程中失去了功能模式 覆盖整个大脑或落在选定的ROI之外。在整个过程中 多年来，科学家们一直在稳步增加预定义ROI的数量 直到它们覆盖整个大脑。然而，由于数量很少， 正电子发射计算机断层扫描研究中的受试者和受试者之间的巨大变异性， 对这些发现的重要性的统计限制开始发挥作用。 与现有的方法相比，我们寻求的客观方法有几个优点。 (1)它将被设计为从每个位置提取显著的几何信息 完整的三维PET图像，如大脑的几何质心和 大脑的主要几何轴，提供了一个参照系 对被摄体头部在PET摄像机内的位置不敏感 不需要额外的X光或MRI图像；(2)它将使用 参照系分析整体所包含的信息 功能图像，而不仅仅是 预先选择感兴趣的解剖区域；(3)它将使用描述性 增强其信噪比的功能图像的特征， 对于确定性、随机噪声和光子散射，即部分 体积效应，典型的从PET或SPECT获得的功能图像 因此，使用较小的相机可以获得良好的对象间平均 受试者数量；(4)它将能够映射检测到的功能 脑图谱或MRI扫描的模式，从而选择解剖模式 后遗症感兴趣的区域。该方法将通过以下两种方式进行验证 幻影大脑和包括休息的PET图像组， 运动激活和光学激活法线。它将进行测试， 一组PET图像，包括正常人和精神分裂症患者。

The overall objective of this research project is the development and validation of a computer-based method that helps in the analysis of functional patterns occurring in 3-dimensional human brain images produced by PET or SPECT cameras. This method is related to health problems since it (1) characterizes patterns of brain dysfunction occurring in mental illness; (2) relates patterns of metabolic activity caused by medication, drug abuse or other stimuli to anatomical areas of the brain; (3) detects pattern differences between two groups; (4) detects subgroups within a heterogeneous group helping in the clinical classification of a single brain. Heretofore, most methods have relied on a priori definitions of anatomical Regions of Interest (ROI) to reduce the amount of functional information to be analyzed, but in the process lose the functional patterns that span the whole brain or fall outside the selected ROI. Throughout the years scientists have steadily increased the number of predefined ROI's until they cover the whole brain. However, due to the small number of subjects in a PET study and the large inter- subject variability, statistical limitations on the significance of the findings come into play. The objective method we seek has several advantages over existing methods. (1) It will be designed to extract salient geometric information from each whole 3-dimensional PET image, such as the brain's geometric centroid and the brain's principal geometric axis, providing a reference frame which is insensitive to the positioning of the subject's head within the PET camera without requiring additional X-ray or MRI images; (2) it will use this reference frame to analyze the information contained in the whole functional image rather than just the functional information contained in a priori selected anatomical regions of interest; (3) it will use descriptive features of the functional image which enhance its signal to noise ratio, both for deterministic, random noise and photon scatter, i.e., partial volume effect, typical of functional images obtained from PET or SPECT cameras; thus, a good intersubject averaging can be obtained with a smaller number of subjects; (4) it will be able to map the detected functional pattern to a Brain Atlas or MRI scan thus selecting a pattern of anatomical regions of interest a posteriori. The method will be validated both with phantom brains and with groups of PET images including resting, motor-activated and optically activated normals. It will be tested with groups of PET images including normals and schizophrenics.