TAFN

Temporal attention fusion network with custom loss function for EEG-fNIRS classification

This study proposes a novel temporal attention fusion network (TAFN) with a custom loss function. The TAFN model incorporates attention mechanisms into its long short-term memory and temporal convolutional layers to accurately capture spatial and temporal dependencies in the EEG–fNIRS data. The custom loss function combines class weights and asymmetric loss terms to ensure precise classification of cognitive and motor intentions while addressing class imbalance issues.

TAFN dataset processing utilities

This repository contains modular code components for processing and analyzing EEG and fNIRS data using the TAFN model.

Files

1. Model.py: Defines a deep learning TAFN model suitable for EEG and fNIRS data.

2. CustomLoss.py: Contains a custom loss function for use with the proposed TAFN model.

3. Data_Augmentation.py: Implements data augmentation techniques to enhance EEG and fNIRS data variability.

4. Features_Extraction_fNIRS.py: Extracts features from fNIRS data.

5. Features_Extraction_EEG.py: Extracts features from EEG data.

6. SignalFraming.py: Segments EEG or fNIRS signals into frames based on duration and overlap.

7. BandPassFiltering.py: Applies a bandpass filter to EEG data within specified frequency ranges.

---

Ensure that all necessary dependencies are installed to use these modules effectively.

If you find this work useful, please cite our article:

Bunterngchit, C., Wang, J., Su, J., Wang, Y., Liu, S., & Hou, Z.-G. (2024). Temporal Attention Fusion Network with Custom Loss Function for EEG–fNIRS Classification. Journal of Neural Engineering. 21, 066016. https://doi.org/10.1088/1741-2552/ad8e86

Publicly available datasets used in the article:

Dataset 1 from Shin et al. https://doi.org/10.1038/sdata.2018.3

Dataset link: https://doi.org/10.14279/depositonce-5830.2

Dataset 2 from Shin et al. https://doi.org/10.1109/TNSRE.2016.2628057

Dataset link: https://doc.ml.tu-berlin.de/hBCI/

Additional Dataset 1 from Buccino et al. https://doi.org/10.1371/journal.pone.0146610

Dataset links: http://dx.doi.org/10.6084/m9.figshare.1619640 and http://dx.doi.org/10.6084/m9.figshare.1619641

Additional Dataset 2 from Shin et al. https://doi.org/10.1371/journal.pone.0196359

Dataset link: https://doi.org/10.6084/m9.figshare.5900842.v1