Segment alignment based cross-subject motor imagery classification under fading data

Area of Science:

• Neuroscience
• Machine Learning
• Brain-Computer Interfaces

Background:

• Motor imagery (MI) utilizes brain imagination for motor learning, with machine learning algorithms like Common Spatial Patterns (CSP) aiding MI signal analysis.
• Conventional MI signal analysis faces challenges in feature extraction and recognition, specifically high personalization across subjects and data fading, which degrades signal quality.
• Data fading is a newly identified issue impacting MI signal integrity, hindering reliable classification and requiring novel solutions.

Purpose of the Study:

• To propose a cross-subject fading data classification approach to address personalization and data fading in motor imagery (MI) signal analysis.
• To enable classification of fading MI data from a single target subject using a model trained on normal data from multiple source subjects.
• To enhance the robustness and accuracy of MI-based brain-computer interfaces (BCIs) despite signal quality variations.

Main Methods:

• A novel cross-subject fading data classification approach incorporating segment alignment is developed.
• The proposed method trains a classification model using normal MI data from multiple subjects.
• The trained model is then applied to classify fading MI data from a single target subject, even under varying fading levels.

Main Results:

• The proposed method demonstrated effective classification performance across different subjects and varying levels of data fading.
• Experimental validation using a BCI Competition dataset and a custom lab-based experiment confirmed the method's efficacy.
• The approach successfully overcomes the limitations of high personalization and data fading in MI signal analysis.

Conclusions:

• The developed cross-subject fading data classification approach with segment alignment effectively addresses key challenges in motor imagery signal processing.
• This method offers a robust solution for classifying fading MI data, improving the reliability of BCIs across diverse users and signal conditions.
• The findings pave the way for more personalized and resilient brain-computer interface applications by mitigating data fading and subject-specific variations.