# Deep learning-based electroencephalic decoding of the phase-lagged transcranial alternating current stimulation

## TL;DR

This study shows that EEG signals can decode the type of brain stimulation applied during a cognitive task, suggesting potential for closed-loop brain-machine interfaces.

## Contribution

A deep learning model successfully decodes phase-lag tACS conditions using task-based EEG signals.

## Key findings

- The model achieved 81.73% decoding accuracy using parietal EEG signals.
- Beta band activity in the parietal region was heavily weighted by the model.
- EEG spectral features may reflect neuromodulatory effects during tACS.

## Abstract

We investigated whether the phase-lag types of cross-frequency coupled alternating current stimulation (CFC-tACS), a non-invasive technique aimed at enhancing cognitive functions, could be decoded using task-based electroencephalographic (EEG) signals. EEG recordings were obtained from 21 healthy individuals engaged in a modified Sternberg task. CFC-tACS was administered online for 6 s during the middle of the retention period with either a 45° or 180° phase lag between the central executive network and the default mode network. To decode different phase-lag tACS conditions, we trained a modified EEGNet using task-based EEG signals before and after the online tACS application. When utilizing parietal EEG signals, the model achieved a decoding accuracy of 81.73%. Feature maps predominantly displayed EEG beta activity in the parietal region, suggesting that the model heavily weighted the beta band, indicative of top-down cognitive control influenced by tACS phase-lag type. Thus, EEG signals can decode online stimulation types, and task-related EEG spectral characteristics may indicate neuromodulatory activity during brain stimulation. This study could advance communicative strategies in brain–machine interfacing (BMI)-neuromodulation within a closed-loop system.