Real-Time Functional MRI with Automated Pattern Classification

描述(申请人提供)：随着功能磁共振成像(FMRI)在神经科学中的快速发展，以及越来越多的研究在临床研究中使用功能磁共振成像，需要自动化的方法来分析和解释数据。功能磁共振成像的高灵敏度使在线分析成为可能，这需要从潜在的感觉、运动和认知过程方面在线解释结果。该功能可以帮助用户最大限度地利用从fMRI获得的信息，并对数据质量做出决策。我们的目标是在样本数据集有限的情况下大幅提高超高维fMRI数据的多类模式分类性能，并开发基于我们定制设计的实时fMRI分析平台(TurboFire)的集成工具，以执行正在进行的实时fMRI扫描期间动态变化的激活模式的模式分类。其目的是(A)开发用于实时模式识别的集成的高性能fMRI分析链，(B)开发用于模式分类的新的稀疏性自适应聚集和偏最小二乘方法，以及(C)通过展示运动、视觉、听觉和心理计算任务的方法来表征数据分析链在分类空间分布的激活模式方面的性能。这一自动、实时功能磁共振成像方法的成功演示将为在正在进行的扫描过程中对动态变化的大脑激活模式进行分类提供概念验证。这将为在临床环境中成功执行感觉和运动功能提供一个标准，并有助于在研究环境中识别更高的认知过程。分类结果的在线显示还将帮助用户调整范式以提高特异性，并使受试者能够进行互动访谈，以进一步探索潜在的大脑过程。该项目的长期目标是开发一种自动化功能磁共振分析工具，该工具将具有巨大的临床和研究使用商业潜力。

DESCRIPTION (provided by applicant): With the rapid development of functional MRI (fMRI) in neuroscience and the increasing number of studies using fMRI in clinical research, there is a need for automated methods to analyze and interpret the data. The high sensitivity of fMRI enables online analysis, which calls for online interpretation of the results in terms of underlying sensory, motor and cognitive processes. This capability can help the user to maximize the information obtained from fMRI and to make decisions on the data quality. The goals are to substantially improve the performance of multi-class pattern classification of very high- dimensional fMRI data with limited number of sample data sets, and to develop an integrated tool based on our custom- designed real-time fMRI analysis platform (TurboFIRE) to perform pattern classification of dynamically changing activation patterns during an ongoing real-time fMRI scan. The aims are (a) Develop an integrated high-performance fMRI analysis chain for real-time pattern recognition, (b) Develop novel sparsity-adaptive aggregation and PLS methods for pattern classification, and (c) Characterize the performance of the data analysis chain for classifying spatially distributed activation patterns by demonstrating the methods on motor, visual, auditory and mental computation tasks. The successful demonstration of this automatic, real-time functional MRI methodology will provide a proof-of-concept of classifying dynamically changing brain activation patterns during the ongoing scan. This will provide a criterion for successfully performing a sensory and motor function in the clinical setting and facilitate the identification of higher cognitive processes in the research setting. Online display of the classification result will also assist the user in adapting the paradigm to improve specificity and enables an interactive interview of the subject to further explore underlying brain processes. The long-term goal of this project is to develop an automated fMRI analysis tool that will have significant commercial potential for clinical and research use.