Incorporation of dynamic stopping strategy into the high-speed SSVEP-based BCIs
Incorporation of dynamic stopping strategy into the high-speed SSVEP-based BCIs
复制标题
将动态停止策略纳入基于 SSVEP 的高速 BCI
DOI：
10.1088/1741-2552/aac605
复制
发表时间：
2018-08-01
影响因子：
4
通讯作者：
Ming, Dong
中科院分区：
文献类型：
--
作者：
Jiang, Jing;Yin, Erwei;Ming, Dong
Objective.脑电是一个非线性、非平稳的过程，其特征不稳定，且在不同的实验中质量不一，这给脑机接口带来了巨大的挑战。这个问题的一个补救措施是自适应地收集足够的EEG证据，使用动态停止（DS）策略。近年来，基于高速稳态视觉诱发电位（SSVEP）的脑机接口取得了巨大的进展。本研究旨在通过引入DS策略进一步改进基于SSVEP的高速脑机接口。Approach.本研究针对一个高速的基于SSVEP的脑机接口，提出了两种不同的DS策略，分别基于贝叶斯估计和判别分析。为了评估它们的性能，我们使用模拟在线测试对我们收集的数据和公共数据集进行了比较，与传统的固定停止（FS）策略。两种最有效的SSVEP识别方法，包括扩展的典型相关分析（CCA）和集成任务相关成分分析（TRCA）被用来进行比较。主要结果。DS策略实现了显着更高的信息传输率（ITR）比FS策略的两个数据集，提高了9.78%的贝叶斯DS和6.7%的判别式DS。具体地，使用集合TRCA的基于判别式的DS策略对于我们收集的数据表现最好，达到353.3 +/-67.1比特min（-1）的平均ITR，峰值为460比特min（-1）。使用集成TRCA的基于贝叶斯的DS策略对于公共数据集具有最高的ITR，达到平均230.2 +/- 65.8 bits min（-1），峰值为304.1 bits min（-1）。意义本研究表明，所提出的动态停止策略可以进一步提高基于SSVEP的脑机接口的性能，并具有实际应用的前景。
Objective. Electroencephalography (EEG) is a non-linear and non-stationary process, as a result, its features are unstable and often vary in quality across trials, which poses significant challenges to brain-computer interfaces (BCIs). One remedy to this problem is to adaptively collect sufficient EEG evidence using dynamic stopping (DS) strategies. The high-speed steady-state visual evoked potential (SSVEP)-based BCI has experienced tremendous progress in recent years. This study aims to further improve the high-speed SSVEP-based BCI by incorporating the DS strategy. Approach. This study involves the development of two different DS strategies for a high-speed SSVEP-based BCI, which were based on the Bayes estimation and the discriminant analysis, respectively. To evaluate their performance, they were compared with the conventional fixed stopping (FS) strategy using simulated online tests on both our collected data and a public dataset. Two most effective SSVEP recognition methods were used for comparison, including the extended canonical correlation analysis (CCA) and the ensemble task-related component analysis (TRCA). Main results. The DS strategies achieved significantly higher information transfer rates (ITRs) than the FS strategy for both datasets, improving 9.78% for the Bayes-based DS and 6.7% for the discriminant-based DS. Specifically, the discriminant- based DS strategy using ensemble TRCA performed the best for our collected data, reaching an average ITR of 353.3 +/- 67.1 bits min(-1) with a peak of 460 bits min(-1). The Bayes-based DS strategy using ensemble TRCA had the highest ITR for the public dataset, reaching an average of 230.2 +/- 65.8 bits min(-1) with a peak of 304.1 bits min(-1). Significance. This study demonstrates that the proposed dynamic stopping strategies can further improve the performance of a SSVEP-based BCI, and hold promise for practical applications.