基于噪声空间不变性的脑电逆问题方法及其在听觉认知中的应用研究

EEG is a kind of non-invasive techniques with millisecond order time resolution and therefore is widely used to dynamically investigate the neural machanism of brain information processing in real time. Imaging the sources of neural activities within brain using scalp EEG is called EEG inverse problem, and still is a hot topic due to its importance and difficulty. Spatio-temporal model-based EEG inverse problem solutions, i.e.,Classical MUSIC and Beamformer, are able to identify EEG sources with high resolution and recover source time courses. However, these methods encounter difficulty in resolving closely spaced sources, highly correlated ones or sources with large amplitude difference. This study systematically analyze signal subspace and noise subspace characteristic of measured EEG correlation matrix, develop noise space invariant based spatio-temporal source localization methods and techniques of reconstruction of source time courses. Primary simulation showed us that we are highly likely to formulate and perfect such methods, build up an effective source localization methods for the cases of closely spaced sources, highly correlated sources or sources with large amplitude difference. We will further apply such methods to analyse the brain sources involved in "where" and "what" dual pathways in auditory cortex of human brain and mutual correlation between these sources, finally try to explain intrinsic neural machanism. At the same time such actual application will in turn facilitate to further perfect the methods and techniques.

脑电图(EEG)具有毫秒量级的时间分辨率、非侵入和无创性等显著特点，被用于实时动态研究大脑信息处理的神经机制。通过头表脑电反演脑内神经活动发生源称为脑电逆问题。该问题的重要性和求解的困难使它仍然是当前脑电领域研究的热点和难点。基于时空模型的脑电逆问题解法，如经典的MUSIC和beamformer，能够用较高的空间分辨率定位脑电源，并导出源的时间过程。然而，这些方法在重建相近源、高度相干源和/或强弱组合源方面有困难。本课题将在系统分析脑电观测数据相关矩阵的信号空间和噪声空间的特性的基础上，发展基于噪声空间不变性的时空脑电源定位和波形重建方法，初步的仿真显示，可望形成一种有效的针对相近源、高度相干源及强弱组合源等情况的定位方法。本工作将进一步把这种方法应用于人脑听觉的"where"和"what"双通路的相关脑区定位和相互作用分析中，解释其内在的神经机制，同时促进对方法技术的发展完善。

项目完成情况：.针对在脑电图和脑电图神经活动源定位过程中存在的困难，我们发展了一种新的方法，基于噪声空间不变性的MEG/EEG源定位算法。通过详尽的真实头模型仿真和真实采集的MEG数据分析，证明新的方法相比传统的方法，能较好地定位相近源，强弱组合源和高度相干源。因为人脑双侧初级听觉皮层波形高度相关，经典的源定位方法在定位初级听觉皮层定位有困难。我们对采集的7个被试的AEF (Auditory Evoked Fields)做了 源分析，结果表明新的方法可以较好地定位这种典型的相干源。该成果发表在PLOS one上，题目为《MEG Source Localization Using Invariance of Noise Space》。针对审稿人提出的，我们的新方法和经典的MUSIC (Multiple Signal Classification ) 之间是否有差异的问题，我们做了详尽的公式推导和仿真实验，这个结果作为论文的附加材料一并发表。.取得成果：.发展了一个新的脑电图脑磁图源定位方法，发表在国际期刊上；对该方法进一步做了分析和研究，将在后继的成果中体现。