Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HCFT: Hierarchical Convolutional Fusion Transformer for EEG Decoding

Haodong Zhang, Jiapeng Zhu, Yitong Chen, Hongqi Li

TL;DR

The paper introduces HCFT, a Hierarchical Convolutional Fusion Transformer designed for EEG decoding that jointly models local temporal/spatiotemporal features and global dependencies across multiple scales. It integrates dual-branch convolutional encoders with cross-branch attention and a hierarchical Transformer fusion to capture fine-grained rhythms and long-range patterns efficiently, with optional Dynamic Tanh normalization for training stability. Evaluations on BCI Competition IV-2b and CHB-MIT show HCFT achieving state-of-the-art or competitive performance (e.g., 80.83% accuracy with kappa 0.6165 on MI data and 99.10% sensitivity with 0.0236/h FPR and 98.82% specificity on seizure prediction), and ablations confirm the critical roles of cross-attention, self-attention, and multi-scale fusion. The work demonstrates strong cross-subject generalization and suggests HCFT’s potential for real-world BCI applications and future extensions toward foundation-model-scale EEG decoding and multimodal integration.

Abstract

Electroencephalography (EEG) decoding requires models that can effectively extract and integrate complex temporal, spectral, and spatial features from multichannel signals. To address this challenge, we propose a lightweight and generalizable decoding framework named Hierarchical Convolutional Fusion Transformer (HCFT), which combines dual-branch convolutional encoders and hierarchical Transformer blocks for multi-scale EEG representation learning. Specifically, the model first captures local temporal and spatiotemporal dynamics through time-domain and time-space convolutional branches, and then aligns these features via a cross-attention mechanism that enables interaction between branches at each stage. Subsequently, a hierarchical Transformer fusion structure is employed to encode global dependencies across all feature stages, while a customized Dynamic Tanh normalization module is introduced to replace traditional Layer Normalization in order to enhance training stability and reduce redundancy. Extensive experiments are conducted on two representative benchmark datasets, BCI Competition IV-2b and CHB-MIT, covering both event-related cross-subject classification and continuous seizure prediction tasks. Results show that HCFT achieves 80.83% average accuracy and a Cohen's kappa of 0.6165 on BCI IV-2b, as well as 99.10% sensitivity, 0.0236 false positives per hour, and 98.82% specificity on CHB-MIT, consistently outperforming over ten state-of-the-art baseline methods. Ablation studies confirm that each core component of the proposed framework contributes significantly to the overall decoding performance, demonstrating HCFT's effectiveness in capturing EEG dynamics and its potential for real-world BCI applications.

HCFT: Hierarchical Convolutional Fusion Transformer for EEG Decoding

TL;DR

The paper introduces HCFT, a Hierarchical Convolutional Fusion Transformer designed for EEG decoding that jointly models local temporal/spatiotemporal features and global dependencies across multiple scales. It integrates dual-branch convolutional encoders with cross-branch attention and a hierarchical Transformer fusion to capture fine-grained rhythms and long-range patterns efficiently, with optional Dynamic Tanh normalization for training stability. Evaluations on BCI Competition IV-2b and CHB-MIT show HCFT achieving state-of-the-art or competitive performance (e.g., 80.83% accuracy with kappa 0.6165 on MI data and 99.10% sensitivity with 0.0236/h FPR and 98.82% specificity on seizure prediction), and ablations confirm the critical roles of cross-attention, self-attention, and multi-scale fusion. The work demonstrates strong cross-subject generalization and suggests HCFT’s potential for real-world BCI applications and future extensions toward foundation-model-scale EEG decoding and multimodal integration.

Abstract

Electroencephalography (EEG) decoding requires models that can effectively extract and integrate complex temporal, spectral, and spatial features from multichannel signals. To address this challenge, we propose a lightweight and generalizable decoding framework named Hierarchical Convolutional Fusion Transformer (HCFT), which combines dual-branch convolutional encoders and hierarchical Transformer blocks for multi-scale EEG representation learning. Specifically, the model first captures local temporal and spatiotemporal dynamics through time-domain and time-space convolutional branches, and then aligns these features via a cross-attention mechanism that enables interaction between branches at each stage. Subsequently, a hierarchical Transformer fusion structure is employed to encode global dependencies across all feature stages, while a customized Dynamic Tanh normalization module is introduced to replace traditional Layer Normalization in order to enhance training stability and reduce redundancy. Extensive experiments are conducted on two representative benchmark datasets, BCI Competition IV-2b and CHB-MIT, covering both event-related cross-subject classification and continuous seizure prediction tasks. Results show that HCFT achieves 80.83% average accuracy and a Cohen's kappa of 0.6165 on BCI IV-2b, as well as 99.10% sensitivity, 0.0236 false positives per hour, and 98.82% specificity on CHB-MIT, consistently outperforming over ten state-of-the-art baseline methods. Ablation studies confirm that each core component of the proposed framework contributes significantly to the overall decoding performance, demonstrating HCFT's effectiveness in capturing EEG dynamics and its potential for real-world BCI applications.
Paper Structure (20 sections, 16 equations, 5 figures, 6 tables)

This paper contains 20 sections, 16 equations, 5 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overall Model Architechture
  4. CFT Block
  5. Dual-Branch Convolutional Feature Extractor
  6. Attention-based Fusion Module
  7. Multi-Scale Aggregation and Final Classification
  8. DATASET AND EXPERIMENTAL SETUP
  9. Datasets and Preprocessing
  10. Experiment Setting
  11. Performance Metrics
  12. Implementation Details
  13. Model Parameters
  14. RESULTS
  15. Head-to-head Comparison Results
  16. ...and 5 more sections

Figures (5)

  • Figure 1: The proposed HCFT framework, include Multiple Stage design with different number of CFT Blocks
  • Figure 2: Accuracy effect of embedding dimension and number of heads.
  • Figure 3: Influence of number of Stage 3 layers on model performance.
  • Figure 4: The proposed HCFT framework, include Multiple Stage design with different number of CFT Blocks
  • Figure 5: The features for subjects ((a)-S4, (b)-S5, (c)-S7, (d)-S8) of Dataset I by t-SNE visualization.