Novel fNIRS study on homogeneous symmetric feature-based transfer learning for brain-computer interface.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCI) enable communication by translating brain activity into external commands.
• Functional Near-Infrared Spectroscopy (fNIRS) is a popular non-invasive method for detecting brain activity.
• Deep learning enhances BCI performance but requires extensive data, long training times, and significant computational resources.

Purpose of the Study:

• Investigate transfer learning as a novel solution for fNIRS-based BCI.
• Address challenges of insufficient training data, extended recalibration periods, and high computational demands.
• Improve accuracy and efficiency in deep learning models for BCI.

Main Methods:

• Applied symmetric homogeneous feature-based transfer learning.
• Utilized a convolutional neural network (CNN) specifically designed for fNIRS data.
• Collected data from 26 participants performing an n-back task.

Main Results:

• The transfer learning model achieved maximum accuracy faster than traditional CNNs.
• Outperformed the traditional CNN model by 25.58% in averaged accuracy over the same training duration.
• Demonstrated reduction in training time, recalibration time, and computational resource requirements.

Conclusions:

• Transfer learning offers a viable solution to overcome deep learning limitations in fNIRS-based BCI.
• The proposed method enhances efficiency and accuracy, making BCI systems more accessible.
• This approach paves the way for more practical and widespread application of deep learning in BCI technology.