A visual working memory dataset collection with bootstrap Independent Component Analysis for comparison of electroencephalogram preprocessing pipelines

Here we present an electroencephalographic (EEG) collection of 71-channel datasets recorded from 14 subjects (7 males, 7 females, aged 20-40 years) while performing a visual working memory task. These data were also used in (Delorme, Palmer, Onton, Oostenveld, Makeig. “Independent EEG sources are dipolar.” PLoS One, 2012). Here, each dataset includes a set of 150 Independent Component Analysis (ICA) decompositions by Extended Infomax using RELICA, each on a bootstrap resampling of the data. Independent components (ICs) are clustered within subject and thereby associated with a quality index (QIc) measure of their stability to data resampling. Sets of single ICA decompositions obtained after applying Principal Component Analysis (PCA) to the data to perform dimension reduction retaining (85%, 95%, 99%) of data variance are also included, as are the positions of the best fitting equivalent dipoles for ICs whose scalp projections are compatible with a compact brain source. These bootstrap ICs may be used as benchmarks for different data preprocessing pipelines and/or ICA algorithms, allowing investigation of the effects that noise or insufficient data have on the quality of ICA decompositions.

本研究提供了一套71通道脑电（EEG，electroencephalographic）数据集，数据采集自14名被试（7名男性、7名女性，年龄区间为20~40岁），采集过程中被试正在执行视觉工作记忆任务。该数据集也曾被用于Delorme、Palmer、Onton、Oostenveld与Makeig于2012年发表在《PLoS One》的论文《Independent EEG sources are dipolar》（《独立脑电信号源为偶极子源》）中。
每份数据集均包含150组独立成分分析 (Independent Component Analysis，ICA) 分解结果，所有分解均通过采用RELICA的扩展信息最大化算法 (Extended Infomax using RELICA) 完成，且每一组分解均基于原始数据的一次自助重采样 (bootstrap resampling) 结果实现。独立成分 (Independent Component，IC) 会在单个被试范围内进行聚类，并由此被赋予一项质量指数 (quality index，QIc)，用于衡量其在数据重采样过程中的稳定性。
此外，本数据集还收录了通过主成分分析 (Principal Component Analysis，PCA) 对原始数据进行降维后得到的多组单ICA分解结果集，降维时分别保留了85%、95%与99%的数据方差；同时还收录了头皮投射与紧凑脑内源兼容的独立成分的最优拟合等效偶极子 (equivalent dipoles) 位置信息。
此类基于自助重采样的独立成分可作为不同数据预处理流水线 (preprocessing pipelines) 与/或ICA算法的基准测试集，用于探究噪声或数据量不足对ICA分解质量所产生的影响。