نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

A Review of Advanced Deep Learning Architectures for Robust Pattern Recognition in EEG Signal Analysis

Subject:

Artificial Intelligence and Brain Mapping

Presenter Author:
Name:	Sahar sadat NazmBojnordi

Author:
Name 1:	Sahar sadat NazmBojnordi - Corresponding - Presenter
Affiliation 1:	Department of Neuroscience and Adiction Studies,school of Advanced Technologies in medicine, Tehran University of Medical Sciences, 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 :

Electroencephalography (EEG) remains a cornerstone of neuroscience research and clinical diagnostics due to its excellent temporal resolution. However, the complex, non-stationary, and high-dimensional nature of EEG signals presents significant challenges for traditional signal processing and shallow machine learning models, which often rely on manual, hypothesis-driven feature extraction. The advent of Deep Learning (DL) has initiated a paradigm shift towards end-to-end learning, yet the field is rapidly evolving beyond initial convolutional (CNN) and recurrent neural network (RNN) models. This review synthesizes the current state-of-the-art, focusing on the transition to more sophisticated architectures like Transformers and Graph Neural Networks (GNNs), and underscores the critical role of Explainable AI (XAI) in bridging the gap between predictive power and neuroscientific insight.

Methods:

A systematic literature search was conducted across PubMed, IEEE Xplore, and ArXiv databases for peer-reviewed articles published between 2020 and 2024. Keywords included "EEG," "deep learning," "Transformer," "Graph Neural Network," "attention mechanism," "spatio-temporal," and "Explainable AI." The review prioritizes studies that either introduce novel architectures for EEG-based classification tasks (e.g., seizure detection, sleep scoring, Brain-Computer Interfaces) or apply XAI techniques to interpret DL models in a neuroscientifically meaningful context.

Results:

Our synthesis reveals a clear trajectory in the complexity and capability of DL models for EEG analysis. While CNNs have proven effective for learning spatial filters from electrode arrangements and RNNs (specifically LSTMs) for capturing short-term temporal dependencies, they face limitations in modeling long-range interactions. The current frontier is defined by two major architectural classes: Transformer-based models: Leveraging self-attention mechanisms, architectures like EEG-Conformer and Brain-BERT have demonstrated state-of-the-art performance by effectively capturing long-range temporal and contextual dependencies within EEG epochs. Their ability to weigh the importance of different time points makes them exceptionally powerful for tasks like sleep stage classification and cognitive load estimation. Graph Neural Networks (GNNs): These models conceptualize the EEG montage as a graph, where electrodes are nodes and their functional or structural relationships are edges. GNNs excel at learning complex spatio-temporal patterns and brain connectivity dynamics, showing immense promise in diagnosing neurological disorders like epilepsy and Alzheimer's by identifying aberrant network topologies. Crucially, the rise of these complex models has been paralleled by the development of XAI methods (e.g., attention maps, GNNExplainer), which provide vital tools to visualize and understand the features driving model predictions, thereby enhancing their clinical and scientific utility.

Conclusion:

The field of AI-driven EEG analysis is rapidly advancing from foundational DL models to more powerful and structurally informed architectures. Transformers and GNNs represent the cutting edge, offering unprecedented capabilities in modeling the intricate dynamics of brain activity. However, the ultimate value of these models in neuroscience hinges on our ability to interpret them. The future lies in the synergistic development of these advanced architectures with robust XAI techniques, transforming them from "black box" classifiers into powerful tools for scientific discovery and reliable clinical decision-making.

Keywords:

EEG, Deep Learning, Pattern Recognition, Transformer, Graph Neural Network, Explainable AI (XAI).