TriDim blocks improve brainwave signal decoding across varied tasks
Beyond Flattened Tokens: Structure-Preserving EEG Decoding with Reusable TriDim Blocks
Machine Learning
Summary
EEG devices record brain signals over time and channels, but analyzing these signals well requires keeping track of space and time without mixing them up. The authors created TriDim blocks that treat EEG data as three connected dimensions—channels, short time chunks, and longer time patches—keeping these parts separate but connected. Using these blocks, their TriDimEEG system better understands brain signals for tasks like diagnosing conditions, sleep analysis, and emotion detection compared to prior methods. This approach also makes existing models more accurate and smaller when the authors swap in TriDim blocks.
What this means in practice
- •For clinical neurotechnology teams: Improve diagnostic EEG tools by using TriDimEEG to better separate spatial and temporal EEG features for more accurate brain disorder detection.
- •For biomedical device engineers: Enhance brain-computer interface systems by integrating TriDim blocks for more efficient and accurate decoding of EEG signals in real time.
Authors
Shiyue Su, Song Wang, Zekai Zhan, Junjie Zeng, Ziling Lu, Zongsheng Li, Xinyuan Ye, Zhiyuan Ma, Xinke Shen, Quanying Liu
Abstract
Effective EEG decoding requires representations that preserve organization among channels, local waveform dynamics, and long-range temporal context. Existing EEG architectures often capture these structures using separate specialized modules or collapse them into a single token sequence, making it difficult to maintain their distinct roles and coordinate their interactions throughout the backbone. We propose TriDim, a reusable block that preserves the representation shape and keeps three EEG axes explicit: channel, sample position within each patch, and patch position across the recording. These axes correspond to spatial, short-term temporal, and long-term temporal information, respectively. Each TriDim block applies feed-forward transformations along individual axes and cross-axis attention to coordinate information exchange among them. By stacking TriDim blocks with a multi-level tri-axis readout, we construct TriDimEEG, a standalone EEG decoder. Under strict cross-subject evaluation on eight datasets spanning clinical diagnosis, sleep staging, motor imagery, and emotion recognition, TriDimEEG achieves the best overall performance among fifteen evaluated models, with a 4.3% relative improvement in average accuracy over the second-best model. Replacing Transformer blocks in three EEG foundation models with TriDim blocks yields an average relative improvement of 7.4% in downstream accuracy while reducing parameter counts by 17.0% to 47.3%. These results establish TriDim as an effective and reusable building block and TriDimEEG as a strong standalone EEG decoder. Code and parameters of TriDimEEG are available at https://github.com/ncclab-sustech/TriDim_model.