Komosar Milana, Tamburro Gabriella, Graichen Uwe, Comani Silvia, Haueisen Jens
Institute of Biomedical Engineering and Informatics at the Technische Universität Ilmenau, Ilmenau, Germany.
Behavioral Imaging and Neural Dynamics Center, G. d'Annunzio University of Chieti-Pescara, Chieti, Italy.
Front Neurosci. 2025 Jun 27;19:1576954. doi: 10.3389/fnins.2025.1576954. eCollection 2025.
Dry electroencephalography (EEG) allows for recording cortical activity in ecological scenarios with a high channel count, but it is often more prone to artifacts as compared to gel-based EEG. Spatial harmonic analysis (SPHARA) and ICA-based methods (Fingerprint and ARCI) have been separately used in previous studies for dry EEG de-noising and physiological artifact reduction. Here, we investigate if the combination of these techniques further improves EEG signal quality. For this purpose, we also introduced an improved version of SPHARA.
Dry 64-channel EEG was recorded from 11 healthy volunteers during a motor performance paradigm (left and right hand, feet, and tongue movements). EEG signals were denoised separately using Fingerprint + ARCI, SPHARA, a combination of these two methods, and a combination of these two methods including an improved SPHARA version. The improved version of SPHARA includes an additional zeroing of artifactual jumps in single channels before application of SPHARA. The EEG signal quality after application of each denoising method was calculated by means of standard deviation (SD), signal to noise ratio (SNR), and root mean square deviation (RMSD), and a generalized linear mixed effects (GLME) model was used to identify significant changes of these parameters and quantify the changes in the EEG signal quality.
The grand average values of SD improved from 9.76 (reference preprocessed EEG) to 8.28, 7.91, 6.72, and 6.15 μV for Fingerprint + ARCI, SPHARA, Fingerprint + ARCI + SPHARA, and Fingerprint + ARCI + improved SPHARA, respectively. Similarly, the RMSD values improved from 4.65 to 4.82, 6.32, and 6.90 μV, and the SNR values changed from 2.31 to 1.55, 4.08, and 5.56 dB.
Our results demonstrate the different performance aspects of Fingerprint + ARCI and SPHARA, artifact reduction and de-noising techniques that complement each other. We also demonstrated that a combination of these techniques yields superior performance in the reduction of artifacts and noise in dry EEG recordings, which can be extended to infant EEG and adult MEG applications.
干式脑电图（EEG）能够在自然场景中以高通道数记录皮层活动，但与基于凝胶的脑电图相比，它更容易产生伪迹。空间谐波分析（SPHARA）和基于独立成分分析（ICA）的方法（指纹法和ARCI）在以往的研究中分别用于干式脑电图的去噪和生理伪迹减少。在此，我们研究这些技术的组合是否能进一步提高脑电图信号质量。为此，我们还引入了SPHARA的改进版本。
在运动表现范式（左手和右手、脚以及舌头运动）期间，从11名健康志愿者身上记录干式64通道脑电图。分别使用指纹法+ARCI、SPHARA、这两种方法的组合以及包含改进版SPHARA的这两种方法的组合对脑电图信号进行去噪。SPHARA的改进版本包括在应用SPHARA之前对单通道中的伪迹跳跃进行额外归零。通过标准差（SD）、信噪比（SNR）和均方根偏差（RMSD）计算每种去噪方法应用后的脑电图信号质量，并使用广义线性混合效应（GLME）模型来识别这些参数的显著变化并量化脑电图信号质量的变化。
对于指纹法+ARCI、SPHARA、指纹法+ARCI+SPHARA和指纹法+ARCI+改进版SPHARA，SD的总体平均值分别从9.76（参考预处理脑电图）提高到8.28、7.91、6.72和6.15 μV。同样，RMSD值从4.65提高到4.82、6.32和6.90 μV，SNR值从2.31变为1.55、4.08和5.56 dB。
我们的结果展示了指纹法+ARCI和SPHARA这两种相互补充的伪迹减少和去噪技术的不同性能方面。我们还证明，这些技术的组合在减少干式脑电图记录中的伪迹和噪声方面具有卓越性能，这可以扩展到婴儿脑电图和成人脑磁图应用中。
