Inter-individual and inter-site neural code conversion without shared stimuli.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Individual differences in brain activity patterns (functional topographies) complicate large-scale data analysis.
• Existing functional alignment methods require identical stimuli across subjects, which is often impractical.
• Developing methods to analyze brain data across individuals without shared stimuli is crucial for scalable neuroscience.

Purpose of the Study:

• To develop a neural code conversion method that aligns brain activity across individuals without requiring shared stimuli.
• To enable accurate decoding and reconstruction of brain activity representations between subjects.
• To establish a foundation for scalable neural data analysis and brain-to-brain communication.

Main Methods:

• A novel neural code conversion approach was developed, optimizing parameters based on stimulus content discrepancies between original and converted brain activity.
• Hierarchical features from deep neural networks were utilized as latent content representations.
• The method was validated by decoding converted brain activity using target subject decoders for image reconstruction.

Main Results:

• The neural code conversion method achieved high accuracy, comparable to methods using shared stimuli.
• Visual image reconstructions from converted brain activity rivaled within-individual decoding quality.
• Successful decoding and reconstruction were demonstrated even with data from different sites and limited training samples.

Conclusions:

• This conversion method effectively overcomes the constraint of shared stimuli for functional alignment.
• The approach provides a robust framework for scalable neural data analysis and modeling.
• The findings lay the groundwork for future brain-to-brain communication applications.