نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Predicting Individual Response to Transcranial Alternating Current Stimulation from Baseline Electroencephalography Signals

Subject:

Personalized Interventions and Diagnosis of Complex Neurological Disorders

Presenter Author:
Name:	Zeinab Jalilzadeh

Author:
Name 1:	Zeinab Jalilzadeh - Corresponding - Presenter
Affiliation 1:	B.Sc. Student, Department of Biomedical Engineering, Amirkabir University of Technology, Tehran, Iran

Name 2:
Affiliation 2:

Name 3:
Affiliation 3:

Name 4:
Affiliation 4:

Name 5:
Affiliation 5:

Name 6:
Affiliation 6:

Name 7:
Affiliation 7:

Name 8:
Affiliation 8:

Name 9:
Affiliation 9:

Name 10:
Affiliation 10:

Abstract

Original Research

introduction :

Individual variability in response to transcranial electrical stimulation (tES) remains an important challenge in non-invasive brain stimulation. Task-baseline electroencephalography (EEG) may provide physiological markers of this variability, but identifying reliable predictors is particularly difficult in small datasets. This study investigated whether task-baseline EEG spectral characteristics could distinguish responders from non-responders to motor-cortex high-definition transcranial alternating current stimulation (HD-tACS).

Methods:

14 EEG recordings from 9 participants were analyzed. The stimulation protocol consisted of 1 mA HD-tACS applied over the motor region at 30 Hz. EEG signals were band-pass filtered between 0.5 and 40 Hz; a notch filter at 60 Hz was applied to remove electrical noise, followed by independent component analysis (ICA), applied to identify and remove physiological artifacts. Power spectral density was estimated to extract frequency-domain features from global and regional brain areas, including absolute and relative band power, power in decibels (dB), and peak alpha and beta frequencies. To mitigate the risk of information leakage inherent in small sample sizes, the analysis was conducted in two phases. In Phase 1, Leave-One-Subject-Out Cross-Validation (LOSO CV) was used to rigorously evaluate the classification pipeline, with normalization and Welch’s t-test feature selection performed strictly within each isolated training fold to train a linear Support Vector Machine (SVM). In Phase 2, mutual information was calculated across the entire dataset to rank the most robust candidate biomarkers for neurophysiological interpretation.

Results:

In Phase 1, the linear SVM successfully classified responders with 78.57% accuracy, 83.33% balanced accuracy, 62.5% precision, 100% recall, and a 76.92% F1-score under strict LOSOCV. In Phase 2, Parietal Theta and Parietal Alpha power (dB) emerged as the strongest discriminative features, followed by Global Beta, Parietal Gamma, and Frontal Beta power. The linear SVM coefficients for Parietal Theta and Parietal Alpha were −0.9014 and −1.1125, respectively.

Conclusion:

The findings suggest that task-baseline EEG spectral characteristics may contain information associated with individual response to motor-cortex tACS. The strict subject-level validation provides a conservative estimate of classification performance while minimizing information leakage. Parietal alpha and theta power emerged as the strongest candidate features, which may reflect the functional state of posterior cortical networks connected with the stimulated motor region. Parietal alpha is associated with sensory processing and attentional regulation, while theta activity is involved in cognitive control and large-scale network interactions. Their negative SVM coefficients indicate that lower baseline parietal alpha and theta power were associated with responder classification in the present dataset. This pattern may reflect differences in the pre-stimulation state of the parieto-motor network and its susceptibility to externally applied oscillatory stimulation, although the underlying mechanism cannot be established from the data. The prominence of parietal rather than exclusively motor features further suggests that individual response to tACS may depend on the broader network state surrounding the stimulation target. Beta and gamma measures provided complementary discriminatory information. However, the small sample of 9 participants limits statistical power and generalizability. Larger independent cohorts are required to determine whether these parietal EEG features can be replicated and validated as predictors of tACS response.

Keywords:

Transcranial Direct Current Stimulation; Electroencephalography; Machine Learning; Biomarkers; Supervised Machine Learning