Conditional-GAN Based Data Augmentation for Deep Learning Task Classifier Improvement Using fNIRS Data.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Functional near-infrared spectroscopy (fNIRS) is an economical, non-invasive neuroimaging technique for mapping cortical activity.
• fNIRS is suitable for population studies and crucial for Brain-Computer Interface (BCI) task classification.
• Deep learning models require substantial data for accurate training, which can be challenging to collect for fNIRS.

Purpose of the Study:

• To improve the accuracy of deep learning classifiers for fNIRS-based BCI task classification.
• To address the challenge of insufficient data for training deep learning models in fNIRS studies.
• To enhance fNIRS data augmentation using generative networks.

Main Methods:

• Utilized Conditional Generative Adversarial Networks (CGAN) for data augmentation of fNIRS signals.
• Integrated a Convolutional Neural Network (CNN) classifier with the CGAN.
• Trained the system to classify tasks: Left Finger Tap, Right Finger Tap, and Foot Tap.

Main Results:

• The CGAN-CNN model achieved a high task classification accuracy of 96.67%.
• Data augmentation via CGAN effectively improved classifier performance with limited fNIRS samples.
• Demonstrated the potential of generative networks in enhancing BCI accuracy.

Conclusions:

• The proposed CGAN-CNN system significantly enhances fNIRS-based BCI task classification accuracy.
• CGANs are effective for augmenting fNIRS data, overcoming sample size limitations.
• This approach offers a promising solution for developing more robust BCIs.