Toward large scale electrophysiology data analysis, many preprocessing pipelines are developed to reject artifacts as the prerequisite step before the downstream analysis. A mainstay of these pipelines is based on the data driven approach -- Independent Component Analysis (ICA). Nevertheless, there is little effort put to the preprocessing quality control. In this paper, attentions to this issue were carefully paid by our observation that after running ICA based preprocessing pipeline: some subjects showed approximately Parallel multichannel Log power Spectra (PaLOS), namely, multichannel power spectra are proportional to each other. Firstly, the presence of PaLOS and its implications to connectivity analysis were described by real instance and simulation; secondly, we built its mathematical model and proposed the PaLOS index (PaLOSi) based on the common principal component analysis to detect its presence; thirdly, the performance of PaLOSi was tested on 30094 cases of EEG from 5 databases. The results showed that 1) the PaLOS implies a sole source which is physiologically implausible. 2) PaLOSi can detect the excessive elimination of brain components and is robust in terms of channel number, electrode layout, reference, and the other factors. 3) PaLOSi can output the channel and frequency wise index to help for in-depth check. This paper presented the PaLOS issue in the quality control step after running the preprocessing pipeline and the proposed PaLOSi may serve as a novel data quality metric in the large-scale automatic preprocessing.

翻译：面向大规模电生理数据分析时，许多预处理流程将伪迹剔除作为下游分析的前置步骤，其中主要方法基于数据驱动方式——独立成分分析(ICA)。然而目前在预处理质量控制的投入十分有限。本文通过观察发现：经过基于ICA的预处理流程后，部分受试者的多通道对数功率谱呈现近似平行状态(PaLOS)，即各通道功率谱相互成比例。首先通过实例与仿真描述了PaLOS现象及其对连接性分析的影响；其次建立其数学模型，基于公共主成分分析提出PaLOS指数(PaLOSi)以检测该现象；最后在5个数据库的30094例脑电数据上测试PaLOSi性能。结果表明：1) PaLOS暗示存在单一来源，这在生理学上不可解释；2) PaLOSi能检测脑成分的过度剔除，对通道数量、电极布局、参考方式等因素具有鲁棒性；3) PaLOSi可输出通道级与频率级指数以辅助深度核查。本文揭示了预处理流程运行后质量控制环节中的PaLOS问题，所提出的PaLOSi可作为大规模自动预处理的创新型数据质量指标。