OBJECTIVE: Recent studies suggest that the use of noninvasive closed-loop neuromodulation combining electroencephalography (EEG) and transcranial alternating current stimulation (tACS) may be a promising avenue for the treatment of neurological disorders. However, the attenuation of tACS artifacts in EEG data is particularly challenging, and computationally efficient methods are needed to enable closed-loop neuromodulation experiments. Here we introduce an original method to address this methodological issue.
APPROACH: Our alternating current regression (AC-REG) method is an adaptive (time-varying) spatial filtering method. It relies on a data buffer of preset size, on which principal component analysis (PCA) is applied. The resulting components are used to build a spatial filter capable of regressing periodic signals in phase with the stimulation. PCA is performed each time that a new sample enters the buffer, such that the spatial filter can be continuously updated and applied to the EEG data.
MAIN RESULTS: The AC-REG accuracy in terms of tACS artifact attenuation was assessed using simulated and real EEG data. Alternative offline processing methods, such as the superimposition of moving averages (SMA) and the Helfrich method (HeM), were used as benchmark. Using simulations, we found that AC-REG can yield a more reliable reconstruction of the stimulation signal for any frequency between 1 and 80 Hz. Analysis of real EEG data of 18 healthy volunteers showed that AC-REG was able to better recover hidden neural activity as compared to SMA and HeM. Also, significantly higher correlations between power spectrum densities in tACS on and off conditions, respectively, were obtained using AC-REG (r=0.90) than using SMA (r=0.80) and HeM (r=0.86).
SIGNIFICANCE: Thanks to its low computational complexity, the AC-REG method can be employed in noninvasive closed-loop neuromodulation experiments, with potential applications both in healthy individuals and in neurological patients.