Neurofeedback of two motor functions using supervised learning-based real-time functional magnetic resonance imaging

Area of Science:

• Neuroscience
• Cognitive Science
• Biomedical Engineering

Background:

• Real-time functional magnetic resonance imaging (rt-fMRI) allows for direct brain activity monitoring.
• Support vector machine (SVM) classification is a machine learning technique applicable to neuroimaging data.
• Understanding neural mechanisms of motor control is vital for clinical applications.

Purpose of the Study:

• To investigate the efficacy of SVM-based rt-fMRI neurofeedback in distinguishing between two distinct motor tasks.
• To explore the neural regions involved in predicting mental states during motor control.
• To examine the impact of active task engagement on brain signal prediction accuracy.

Main Methods:

• Healthy volunteers (n=13) underwent rt-fMRI while performing a button tapping task and a speech counting task.
• SVM classification was employed to predict task performance based on BOLD signals.
• Offline analysis and group spatial map analysis were conducted to assess classification accuracy and identify neural patterns.

Main Results:

• High online prediction accuracies were achieved: ~95% for button tapping and ~86% for speech counting.
• Offline analysis revealed significantly lower initial classification accuracies: 75% for button tapping (p<0.001) and 72% for speech counting (p<0.005).
• Group analysis of SVM spatial maps showed significant differences between tasks, suggesting distinct neural representations.

Conclusions:

• Active engagement in motor tasks, particularly when controlling an interface, enhances BOLD signal intensity and spatial extent.
• SVM-based rt-fMRI neurofeedback can differentiate between motor tasks, with accuracy influenced by task engagement.
• The findings contribute to understanding neural dynamics in motor control and potential applications for brain-computer interfaces.