Exploring Effects of Mental Stress with Data Augmentation and Classification Using fNIRS.

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Accurate brain state identification is crucial in functional brain imaging.
• Small participant numbers limit machine learning classification in brain state research.
• Stress impacts spatial activation and connectivity, measurable via fNIRS.

Purpose of the Study:

• To classify stress-induced brain states using fNIRS data.
• To enhance classification accuracy through data augmentation with deep convolutional generative adversarial networks (DCGAN).
• To investigate the effect of binaural beats (BBs) on stress reduction.

Main Methods:

• Employed a classification strategy on fNIRS data from 21 participants during the Stroop color-word task (SCWT).
• Utilized DCGAN for data augmentation to increase dataset size.
• Applied various machine learning algorithms, including LDA and CNN, for classification.
• Introduced binaural beats (16 Hz) for stress mitigation.

Main Results:

• CNN classification accuracy reached 73% with non-augmented data.
• DCGAN data augmentation dramatically improved classification accuracy to 96%.
• Statistically significant improvements in brain state were observed after BB stimulation.

Conclusions:

• fNIRS can detect brain state changes with high accuracy.
• Data augmentation is effective in overcoming data scarcity for brain signal analysis.
• Binaural beats show potential for stress reduction and improving brain states.