Online MEG Source Localization using High-Performance GPU Computing (OSL)

基于单次脑磁图（MEG）的人脑神经元源的在线定位是近年来神经科学研究的热点之一，其目的在于阐明脑功能、识别脑状态以及提高脑机接口（BCI）系统等实时应用的能力。通过利用图形处理单元（GPU）和NVIDIA计算引擎CUDA，我们的目标是开发高性能的优化源定位算法，允许执行在线源定位典型大小的铅场矩阵和嘈杂的单一试验数据。首先，对RAP-MUSIC算法、MNE算法和Beamformer算法进行了优化和并行化。其次，我们将在上述算法的基础上发展一种新的多层混合逆算法，以减少计算量并处理信噪比不足的情况。第三，我们将包括动态的先验信息，以提高在线定位的准确性和速度。我们将提供一个易于使用的软件框架和开发的算法在一个开放的图书馆，这将是一个至关重要的要求在线大脑状态分类在我们的合作伙伴项目（建议“状态引导的感知”，SGP）。患者将直接受益于新技术，以改善神经植入物参数的适应性;通过状态引导学习中的在线反馈和脑计算机接口中的更高分类率来改善康复结果。

Online localization of neuronal sources in the human brain from single trial magnetoencephalography (MEG) recently got into focus of neuroscience to clarify brain functionality, identify brain states and to improve real-time applications such as brain-computer interface (BCI) systems. By utilizing the graphics processing unit (GPU) and NVIDIAs computing engine CUDA, we aim to develop high performance optimized source localization algorithms allowing to perform online source localization on typically sized lead field matrices and noisy single trial data. First, we will optimize and parallelize the algorithms RAP-MUSIC, MNE and Beamformer. Secondly, we will develop a new multilayer hybrid inverse algorithm based on the algorithms mentioned above to reduce the computational effort and handle insufficient signal-to-noise ratio (SNR). Thirdly, we will include dynamic a priori information to increase accuracy and speed of online localization. We will provide an easy-to-use software framework and the developed algorithms in an open library, which will be a crucial requirement for online brain state classification in our partner project (proposal 'State-guided Perception', SGP). Patients will directly benefit from the new technology to improve adaptations of parameters of neuroimplants; improved rehabilitation outcome through online feedback in state guided learning and higher classification rates in brain computer interfaces.