TopoTempNet: A High-Accuracy and Interpretable Decoding Method for fNIRS-Based Motor Imagery.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Functional near-infrared spectroscopy (fNIRS) is a safe and portable neuroimaging technique suitable for brain-computer interface (BCI) applications, especially for motor imagery (MI) decoding.
• Challenges in fNIRS data analysis include low sampling rates and hemodynamic delays, which hinder accurate temporal modeling and dynamic brain network analysis.
• Existing methods struggle with static graph modeling and interpreting fused features, limiting the performance and interpretability of BCI systems.

Purpose of the Study:

• To propose TopoTempNet, an innovative topology-enhanced temporal network designed to overcome the limitations of fNIRS in biomedical signal decoding.
• To improve the temporal dynamics modeling, static graph analysis, and feature fusion interpretability for fNIRS-based BCI.
• To achieve high-performance and interpretable decoding of brain signals using fNIRS.

Main Methods:

• TopoTempNet integrates multi-level graph features with temporal modeling, incorporating local and global functional connectivity metrics.
• A graph-modulated attention mechanism, combining Transformer and Bi-LSTM, is employed for dynamic modeling of crucial brain connections.
• A multimodal fusion strategy combines raw fNIRS signals, graph structures, and temporal representations into a high-dimensional space for enhanced discrimination.

Main Results:

• TopoTempNet achieved superior decoding accuracy, reaching up to 90.04% ± 3.53%, and improved Kappa scores across three public fNIRS datasets (MA, WG, UFFT).
• Receiver Operating Characteristic (ROC) curves and t-distributed Stochastic Neighbor Embedding (t-SNE) visualizations demonstrated excellent feature discrimination and structural clarity.
• Statistical analysis of graph features highlighted the model's capability to capture task-specific functional connectivity patterns, thereby increasing the interpretability of decoding results.

Conclusions:

• TopoTempNet offers a novel and effective approach for enhancing the performance and interpretability of fNIRS-based BCI systems.
• The topology-enhanced temporal network successfully addresses the challenges posed by fNIRS data, including temporal dynamics and feature fusion.
• This work paves the way for developing more robust and understandable BCI applications leveraging the advantages of fNIRS technology.