Enhancing fMRI Decoded Neurofeedback with Co-adaptive Training: Simulation and Proof-of-principle Evidence

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Neurofeedback training, particularly fMRI-based decoded neurofeedback (DecNef), faces challenges in participants learning to control specific brain patterns.
• Discrepancies between decoder training data and real-time neurofeedback data, including noise and differing contexts, contribute to learning difficulties.

Purpose of the Study:

• To develop and validate a co-adaptation procedure to enhance participant performance in DecNef training.
• To improve the precision and reliability of DecNef protocols for targeting specific brain representations.

Main Methods:

• Developed a co-adaptation procedure using standard machine learning algorithms with a real-time adaptive decoder.
• Tested the procedure using simulations on a previous DecNef dataset.
• Validated the co-adaptation approach with real-time fMRI data from DecNef training sessions.

Main Results:

• Simulations demonstrated that decoder co-adaptation significantly improves performance during neurofeedback training.
• Drift analysis confirmed the stability of the co-adapted decoder throughout training sessions.
• Real-time fMRI data provided proof-of-concept evidence that co-adaptation enhances participants' ability to induce target brain states.

Conclusions:

• Personalized decoders created through co-adaptation can enhance the effectiveness of DecNef training protocols.
• This approach offers improved precision and reliability for targeting specific brain representations, with potential translational research applications.
• The developed tools are openly available to the scientific community.