Real-Time Classification of Motor Imagery Using Dynamic Window-Level Granger Causality Analysis of fMRI Data.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery (MI) involves imagining movements without physical execution.
• Decoding MI from brain signals is crucial for brain-computer interfaces (BCIs).
• Functional magnetic resonance imaging (fMRI) offers high spatial resolution for brain activity but faces challenges in real-time decoding.

Purpose of the Study:

• To develop and validate a novel method for extracting neural signal features from fMRI data to distinguish between left- and right-hand grasping imagination.
• To investigate the utility of effective brain connections, calculated using Dynamic Window-level Granger Causality (DWGC), as features for real-time MI classification.
• To compare the performance of the proposed method with traditional multivoxel pattern classification analysis.

Main Methods:

• An fMRI experiment was conducted to identify brain regions activated during MI.
• Dynamic Window-level Granger Causality (DWGC) was used to calculate effective connections between regions of interest (ROIs).
• A real-time fMRI (rt-fMRI) classification system was developed on the Open-NFT platform, employing Support Vector Machine (SVM) for three-class classification (rest, left hand, right hand).

Main Results:

• The SVM classifier achieved a maximum accuracy of 69.3% for real-time three-class classification using effective connection features.
• This accuracy was, on average, 3% higher than that achieved by traditional multivoxel pattern classification analysis.
• The proposed method demonstrated significant improvements in classification accuracy during the initial stages of MI tasks and reduced latency effects.

Conclusions:

• Effective connections derived from DWGC are valuable features for real-time decoding of motor imagery using fMRI.
• These effective connection features exhibit higher sensitivity to changes in brain states compared to traditional methods.
• The study provides theoretical support and technical guidance for feature extraction in fMRI-based BCI research.