نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Data-Driven Functional Brain Parcellation from Resting-State EEG Using Self-Supervised Learning

Subject:

Artificial Intelligence and Brain Mapping

Presenter Author:
Name:	Morteza Shahedifar

Author:
Name 1:	Morteza Shahedifar - Corresponding - Presenter
Affiliation 1:	Electrical and Computer Eng. University of Tabriz, Iran

Name 2:	Alireza Siadatan
Affiliation 2:	Associate Professor, Senior Member of IEEE, University of Toronto, Canada

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 :

Electroencephalography (EEG) provides a noninvasive means to study brain activity with high temporal resolution, making it a valuable tool for exploring functional organization. Traditional EEG analyses often rely on predefined frequency bands or manual region selection, which can limit the discovery of new or unexpected patterns of neural activity. Recent advances in self-supervised learning (SSL) offer the opportunity to extract meaningful representations from EEG data without relying on external labels. By applying SSL to resting-state EEG, it becomes possible to identify coherent functional units in the brain, providing a data-driven approach to functional parcellation. Such methods can contribute to a better understanding of spatial and temporal organization of brain activity and complement existing neuroimaging approaches.

Methods:

In this study, resting-state EEG recordings from publicly available datasets, including PhysioNet and BCI Competition archives, were analyzed. Preprocessing steps included band-pass filtering in canonical frequency bands, artifact rejection, and normalization. A self-supervised learning framework was implemented to learn representations for each EEG channel over time. Two complementary SSL approaches were explored: (1) contrastive learning, in which temporally augmented segments of the same channel are treated as positive pairs, and (2) temporal predictive modeling, where the model predicts future segments based on past activity. The learned embeddings were clustered to define functional parcels, and cluster stability was assessed using repeated runs and evaluation metrics such as silhouette score and inter-cluster consistency. All analyses were conducted in Python using MNE and PyTorch libraries, enabling reproducibility and extensibility.

Results:

The SSL framework produced coherent clusters corresponding to groups of channels with similar temporal and spectral properties. Across multiple runs and datasets, these functional parcels demonstrated high stability, suggesting that the learned representations capture intrinsic organization in resting-state EEG. Spatial inspection revealed clusters that align with known cortical regions associated with sensorimotor, frontal, and parietal networks. Temporal analysis further highlighted differences in oscillatory behavior between clusters, supporting the functional relevance of the discovered parcels. Comparison with conventional band-based and manual region approaches indicated that the SSL-based parcellation provides a more flexible and data-driven perspective while remaining physiologically interpretable.

Conclusion:

Self-supervised learning provides an effective and reproducible framework for functional brain parcellation from resting-state EEG. By extracting data-driven embeddings and clustering them into stable parcels, this approach uncovers spatial and temporal patterns of brain activity without relying on task-based labels or manual assumptions. The results demonstrate that SSL can complement existing EEG analysis techniques and offer a principled method for exploring functional brain organization. This framework can be readily applied to other publicly available EEG datasets and serves as a foundation for future research in data-driven brain mapping.

Keywords:

EEG;Brain Mapping;Functional Connectivity;Self-Supervised Learning;Neural Networks