基于Lp(0<p≤1)范数的时变脑电分析方法及其在情绪机制、状态评估中的应用研究

揭示情绪加工的动态变化规律，能够为精神疾病的防治提供科学的理论依据，促进认知神经科学等相关领域的深度发展。脑电具备毫秒级的时间分辨率，可以有效记录大脑活动的瞬息万变，从而客观反映个体情绪状态的实时变化。但是，由于眼电伪迹等干扰的存在，现有的脑电分析方法仍然无法摆脱繁琐复杂的伪迹去除环节，临床和认知科学研究依然缺乏可靠的实时情绪分析技术。研究表明Lp(0<p≤1)范数空间能够抑制眼电伪迹和离群值干扰的影响，因此本项目拟基于Lp(0<p≤1)范数发展能够自动抑制伪迹干扰的时变脑电分析方法，为情绪转化的机制研究提供可靠的时频分析和时变脑网络估计方案。在此基础上，利用长短时记忆网络兼顾结构信息捕捉和时变信息保留的双重优势，深度挖掘脑电在时频拓扑和时变网络连接中的动态互补关系，估计互补信息在情绪维度空间的分布，实现可靠的实时情绪分析，为揭示情绪加工的动态变化规律提供科学的依据。

Revealing dynamic changes of emotion processing provides theoretical basis for mental illness prevention and control, and promotes the cognitive neuroscience researches to develop in depth. Electroencephalography (EEG) with millisecond-scale temporal resolution efficiently records the transient brain activity, which can objectively reflect the real-time change of emotional states. However, traditional EEG analysis methods cannot skip the trivial artifacts removal stages due to the influence of ocular artifacts, resulting in lack of real-time emotional analysis methods in both clinical and neuroscience researches. A series of studies converge on a consistent conclusion that the influence of ocular artifacts and outliers can be alleviated in the Lp(0<p≤1) norm space. Thus, this project proposes to develop Lp(0<p≤1) norm-based time-varying EEG analysis, aimed at offering robust time-frequency analysis and dynamic brain networks estimation for the research of emotion processing mechanism. On this basis, it further utilizes superiority of long-short term memory network, which allows for preservation of the local structure and time-varying information extraction, to mine the dynamic complementary relationships between EEG based time-frequency topology and time-varying brain networks in depth, and project them into emotion-dimension space for real-time emotion analysis. Finally, this research will offer scientific evidence to reveal the dynamic changes of emotion processing.