CAREER: Large-Scale Computational Neuroimaging of Brain Electrical Activity

目前基于脑电/脑磁图的神经成像技术在功能神经成像方面受到的关注有限，因为它们在复杂的大脑任务中的可靠性能尚未被成功证明。迫切需要更可靠、更有效、更创新的、具有高空间和时间分辨率的脑电/脑磁图神经成像技术。这项研究的目标是通过整合不同学科的新技术，即多分辨率数学模型、大规模计算、先进的信号和图像处理以及高端测量设备，推动功能神经成像技术在人脑功能研究中的应用。智力优势：建议的研究包括将成熟的L-1范数正则化技术应用于不同的应用领域，即脑电/脑磁图分析。图像分析问题是一个未确定的问题，因此需要某种形式的正则化。传统的方法是应用L-2范数最小化，当测量误差为高斯随机变量时，这是一种有效的方法。然而，当测量误差服从指数分布时，使用L-1范数最小化更有意义。此外，L-1最优解在有限(通常较少)的数据点上得到支持，从而自然地利用了优化问题的稀疏性。本研究的创新之处在于对神经行为稀疏性诱导模型的研究。就神经成像而言，从生物学因素可以看出，在任何一毫秒的时间间隔内，大脑中只有一小部分神经元会“起火”。因此，拟议的研究有可能确定描述这种行为的参数。作为这一点的部分证据，PI已经根据合成数据获得了一些初步结果，当正确答案已知时，结果是令人鼓舞的。广泛的影响：该项目将利用先进的工程原理为科学问题产生新的问题解决策略，并通过减少生命、工作和收入损失的潜力应用于神经科患者的临床实践。通过融入真实的学习教育活动，我们将扩大各级学生、K-12教师和其他教育工作者对生物医学研究和研究教育的参与。

Current EEG/MEG based neuroimaging technologies receive limited attentions in functional neuroimaging because their reliable performances have not been successfully demonstrated in sophisticated brain tasks. The need for more reliable, efficient, innovative EEG/MEG neuroimaging technologies that have high spatial and temporal resolutions is urgent. The goal of proposed research is to advance functional neuroimaging technologies in studying human brain functions, via integrating novel techniques from different disciplines, i.e. multi-resolution mathematic models, large-scale computation, advanced signal and image processing, and high-end measurement devices. Intellectual Merit: The proposed research consists of applying the well-established L-1 norm regularization technique to a different application domain, namely EEG/MEG image analysis. The problem of image analysis is an undetermined problem, and as such some form of regularization is required. The traditional approach is to apply L-2 norm minimization, which is a valid method when measurement errors are Gaussian random variables. However, when the measurement errors are exponentially distributed, it is more meaningful to use L-1 norm minimization. Moreover, the L-1 optimal solution is supported on a finite (usually small) number of data points, thus naturally taking advantage of the sparse nature of the optimization problem. The novelty of the proposed research is the investigation of sparsity inducing models of neural behavior. In the case of neuro-imaging, it is known from biological factors that during any one millisecond interval, only a tiny fraction of the neurons in the brain "fire". The proposed research therefore has the potential to identify the parameters that describe this behavior. As partial evidence of this, the PI has obtained some preliminary results based on synthetic data, when the correct answer is known, and the results are encouraging.Broader Impact: The project will generate novel problem-solving strategies for scientific problems using advanced engineering principles, and to clinical practice for neurological patients through the potential for reduction in loss of life, job, and income. By incorporating authentic learning education activities, we will broaden the participation of all levels of students, K-12 teachers, and other educators, in biomedical research and research educations.