LSTM based GAN Networks for Enhancing Ternary Task Classification Using fNIRS Data.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain activation patterns are unique to specific tasks.
• Functional near-infrared spectroscopy (fNIRS) is a cost-effective, non-invasive neuroimaging technique for mapping brain activity.
• fNIRS data complexity is well-suited for deep learning classification.

Purpose of the Study:

• To enhance brain-computer interface (BCI) accuracy using limited fNIRS data.
• To develop a deep learning model for classifying tasks based on fNIRS patterns.
• To address data scarcity challenges in neuroimaging research.

Main Methods:

• Utilized a Conditional Generative Adversarial Network (CGAN) integrated with Long Short-Term Memory (LSTM) networks.
• Employed data augmentation with GAN-generated synthetic fNIRS data.
• Trained an LSTM classifier to distinguish between Left Finger Tap, Right Finger Tap, and Foot Tap tasks.

Main Results:

• Achieved a classification accuracy of 90.2% for task identification using the LSTM-based GAN approach.
• Demonstrated the effectiveness of synthetic data generation in improving classifier performance.
• Showcased the potential of deep learning and GANs in overcoming data limitations in fNIRS studies.

Conclusions:

• Conditional Generative Adversarial Networks combined with LSTM classifiers significantly enhance fNIRS-based task classification accuracy.
• GAN-based data augmentation is a viable solution for data-scarce neuroimaging applications.
• The proposed model offers a promising avenue for improving BCI systems and clinical diagnostics.