Scholarlink

Transformer-based deep learning models have rapidly gained traction in electroencephalography (EEG) research due to their capacity for modeling long-range temporal dependencies and spatial patterns. This systematic review surveys 201 papers published between 2019 and 2024, with a focus on transformer and hybrid transformer architectures for EEG signal decoding across tasks such as emotion recognition, motor imagery, and attention classification. We categorize key model innovations, including spatial-temporal attention mechanisms, CNN-transformer hybrids, and neural architecture search techniques. Emerging trends highlight the dominance of hybrid models and increasing exploration of pretrained backbones. We also identify methodological gaps in generalization, interpretability, and task-specific benchmarking. To guide future work, we synthesize recommended models and review papers, and propose directions for quantitative meta-analysis and open-source resource development.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Review of Transformer-Based and Hybrid Deep Learning Approaches for EEG Analysis

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Aniket Konkar,
  • Xiaodong Qu

摘要

Transformer-based deep learning models have rapidly gained traction in electroencephalography (EEG) research due to their capacity for modeling long-range temporal dependencies and spatial patterns. This systematic review surveys 201 papers published between 2019 and 2024, with a focus on transformer and hybrid transformer architectures for EEG signal decoding across tasks such as emotion recognition, motor imagery, and attention classification. We categorize key model innovations, including spatial-temporal attention mechanisms, CNN-transformer hybrids, and neural architecture search techniques. Emerging trends highlight the dominance of hybrid models and increasing exploration of pretrained backbones. We also identify methodological gaps in generalization, interpretability, and task-specific benchmarking. To guide future work, we synthesize recommended models and review papers, and propose directions for quantitative meta-analysis and open-source resource development.