Hsu Sheng-Hsiou, Mullen Tim, Jung Tzyy-Ping, Cauwenberghs Gert
Annu Int Conf IEEE Eng Med Biol Soc. 2014;2014:3845-8. doi: 10.1109/EMBC.2014.6944462.
Online Independent Component Analysis (ICA) algorithms have recently seen increasing development and application across a range of fields, including communications, biosignal processing, and brain-computer interfaces. However, prior work in this domain has primarily focused on algorithmic proofs of convergence, with application limited to small `toy' examples or to relatively low channel density EEG datasets. Furthermore, there is limited availability of computationally efficient online ICA implementations, suitable for real-time application. This study describes an optimized online recursive ICA algorithm (ORICA), with online recursive least squares (RLS) whitening, for blind source separation of high-density EEG data. It is implemented as an online-capable plugin within the open-source BCILAB (EEGLAB) framework. We further derive and evaluate a block-update modification to the ORICA learning rule. We demonstrate the algorithm's suitability for accurate and efficient source identification in high density (64-channel) realistically-simulated EEG data, as well as real 61-channel EEG data recorded by a dry and wearable EEG system in a cognitive experiment.
在线独立成分分析（ICA）算法近来在包括通信、生物信号处理和脑机接口等一系列领域得到了越来越多的发展和应用。然而，该领域先前的工作主要集中在算法收敛性证明上，其应用仅限于小型“玩具”示例或相对低通道密度的脑电图（EEG）数据集。此外，适用于实时应用的高效计算在线ICA实现的可用性有限。本研究描述了一种优化的在线递归ICA算法（ORICA），采用在线递归最小二乘（RLS）白化，用于高密度EEG数据的盲源分离。它作为一个具有在线能力的插件在开源BCILAB（EEGLAB）框架内实现。我们进一步推导并评估了对ORICA学习规则的块更新修改。我们证明了该算法适用于在高密度（64通道）逼真模拟的EEG数据以及在认知实验中由干式可穿戴EEG系统记录的真实61通道EEG数据中进行准确高效的源识别。
