Leveraging shared connectivity to aggregate heterogeneous datasets into a common response space.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Data Science

Background:

• Estimating shared neural response spaces across disparate functional magnetic resonance imaging (fMRI) datasets is challenging.
• Existing methods struggle to integrate heterogeneous data while accounting for individual functional topographies.

Purpose of the Study:

• To develop and evaluate a connectivity-based shared response model for unifying fMRI data.
• To improve cross-subject analysis and information dimensionality within a shared neural space.
• To enable robust semantic encoding models across diverse datasets.

Main Methods:

• Factorization of aggregated fMRI datasets into a shared connectivity space and subject-specific transformations.
• Projection of individual response time series into the estimated shared space.
• Evaluation using heterogeneous, naturalistic fMRI data from spoken story listening tasks.

Main Results:

• Projecting data into the shared space significantly improved between-subject classification of story segments.
• The dimensionality of shared information across subjects increased substantially.
• Generalizable improvements were observed for subjects and stories not used in model estimation.
• Shared semantic encoding models showed enhanced performance.

Conclusions:

• The connectivity-based shared response model effectively creates a consensus neural space from diverse fMRI datasets.
• This approach leverages shared connectivity patterns to resolve individual differences and enhance cross-subject comparability.
• The method offers a powerful tool for analyzing complex, naturalistic brain data.