Independent Component Analysis Based Support Vector Machine Classification Method

描述(申请人提供)：功能磁共振成像(FMRI)技术为成瘾领域提供了一种独特的、具有深远价值的能力来探索人类受试者的大脑机制。尽管我们对神经机制的理解取得了进展，但神经成像方法的诊断作用还没有被认识到。此外，我们评估神经网络随时间变化的反应性的能力仍然有限。为了探索新的方法来促进我们对物质使用障碍的理解和诊断，这项提议将利用新的和强大的分析方法来评估现有的可卡因成瘾者和健康对照的功能磁共振成像(FMRI)数据集。PI是药物滥用领域的新成员，将把其他调查领域开发的分析方法引入成瘾领域。该项目的广泛目标是使用功能磁共振数据检查可卡因成瘾受试者大脑功能的差异，并揭示涉及去抑制和决策任务的区域之间的功能连接。这项研究将利用NIDA资助的一项正在进行的研究“冲动、神经缺陷和可卡因成瘾”的数据。本研究评估了在运动去抑制任务和反应逆转的决策任务中的大胆反应，以及在休息时的基本连通性测量。将使用基于数据驱动的熵的算法--独立分量分析(ICA)来提取在成瘾和对照受试者中显示差异的空间独立区域(目标1)。我们假设，所提出的方法将识别特定的体素组(不一定在空间上相连)，用于分离处于去抑制和决策任务的两组。为了揭示这些不同区域之间的功能连通性如何随着时间的推移而变化，我们将使用时间进程(目标2)上的不同窗口大小执行动态(短时间变化)动态功能网络连通性(DFNC)分析。这些发现有望揭示可卡因成瘾人群和健康对照人群中ICA衍生网络变化之间的时间差异。由于第一个目标将发现显著区分体素的云，我们将使用机器学习算法，支持向量机(支持向量机)，通过使用受试者的fMRI数据(目标3)，自动将受试者映射到可卡因成瘾者或对照组之一。我们假设，基于支持向量机的主题分类工具将在两组之间的选择中提供高灵敏度。该框架潜在地提供了一种有用的临床诊断工具，并展示了两组之间关键的脑神经差异。最后一个目的是揭示两个特定的相关解剖区域之间的动态连接，例如近侧前额叶皮质(PFC)、下额叶皮质、岛叶皮质和前扣带回(用于去抑制任务)和眶前叶皮质、背外侧前额叶皮质和前扣带回(用于决策任务)。我们假设，在两组关于两个任务的交互模式中将观察到明确的边界。首次使用非常强大的数据分析工具(ICA)研究可卡因成瘾者和对照组脑部区域的对比，并将其用于支持向量机的分类框架。该提案将把两种极具创新性的技术(DFNC和支持向量机)和一位有前途的初级调查员结合在一起，进入成瘾领域。利用预先存在的数据库为临床上重要的、令人困惑的和现场长期存在的问题提供了一种高成本效益的方法。 与公共卫生相关：兴奋剂的滥用和依赖是一个重大的公共卫生问题，具有巨大的健康、法律、社会和职业相关成本。虽然以生物学为导向的研究，特别是那些使用成像技术的研究，在我们对与成瘾的形成和持续相关的大脑活动的理解方面取得了重大进展，但这些知识对这些疾病的诊断几乎没有帮助。这项研究将把其他研究领域使用的统计程序应用于从可卡因成瘾受试者获得的成像数据，以评估大脑网络的变化，并确定可卡因成瘾患者是否可以与使用可卡因的受试者区分开来。

DESCRIPTION (provided by applicant): Functional MRI (fMRI) techniques have provided the addiction field with a unique and profoundly valuable ability to explore brain mechanisms in human subjects. Despite the resultant advances in our understanding of neural mechanisms, the diagnostic utility of neuroimaging methods has not been realized. In addition, our ability to assess the reactivity of neural networks over time remains limited. To explore new approaches to further our understanding and diagnosis of substance use disorders, this proposal will utilize novel and powerful analytic methods to assess an existing functional magnetic resonance imaging (fMRI) data set in cocaine-addicted subjects and healthy controls. The PI, new to the substance abuse field, will bring analytic approaches developed in other areas of investigation to the field of addiction. The broad goal of this project is to examine differentiations of brain functions in cocaine-addicted subjects using fMRI data and to reveal functional connectivity between regions involving disinhibition and decision-making tasks. This study will exploit data from an ongoing NIDA-funded study, "Impulsivity, Neural Deficits and Cocaine Addiction." This study assesses BOLD response during a motor disinhibition task and a decision- making task of response reversal as well as basal measures of interconnectivity measures during rest. A data driven entropy-based algorithm, independent component analysis (ICA), will be used to extract spatially independent regions that show differences in addicted and control subjects (Aim 1). We hypothesize that the proposed method will identify certain groups of voxels (not necessarily spatially connected) serving separation of two groups who are under disinhibition and decision-making tasks. To uncover how the functional connectivity changes over time between those differentiated regions, we will perform dynamic (changes over short time) dynamic functional network connectivity (DFNC) analysis using various window sizes on time course (Aim 2). These findings are expected to reveal a temporal distinction between ICA-derived network changes in cocaine-addicted and healthy control populations. Since the first aim will find significantly discriminative clouds of voxels, we will use a machine-learning algorithm, the support vector machines (SVM), to automatically map a subject into one of the groups, cocaine-addicted or control, by using subject's fMRI data (Aim 3). We hypothesize that SVM based subject classification tool will provide high sensitivity in selection between the two groups. The framework potentially offers a useful clinical diagnostic instrument as well as demonstrating key brain neural differences between the groups. The last aim is designed to reveal dynamic connectivity between pairs of given anatomic regions of relevance, such as the mesial prefrontal cortex (PFC), inferior PFC, insular cortex, and anterior cingulate (for the disinhibition task) and the orbitofrontal cortex, dorsolateral prefrontal cortex, and anterior cingulate (for the decision-making task). We hypothesize that a clear boundary will be observed in interaction patterns of two groups regarding two tasks. The contrast between brain regions in fMRI data of cocaine-addicted and control subjects will be investigated for the first time using very powerful data analysis tool (ICA) and will be used in classification framework of SVM. The proposal will bring together two extremely innovative techniques (DFNC and SVM) and a promising junior investigator into the field of addiction. The utilization of a pre-existing database offers a highly cost-effective approach to a clinically important, perplexing and persistent problem in the field. PUBLIC HEALTH RELEVANCE: The abuse and dependence of stimulants is a major public health problem with significant health, legal, social and occupational associated costs. While biologically-oriented studies, particularly those using imaging techniques, have produced significant advances in our understanding of brain activity relevant to the development and persistence of addiction, this knowledge has done little to assist in the diagnosis of these disorders. This study will bring statistical procedures used in other fields of research to imaging data obtained from cocaine-addicted subjects to assess changes in brain networks and determine whether the cocaine-addicted patients can be distinguished from non-cocaine using subjects.