Finding Distributed Needles in Neural Haystacks.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• The human brain encodes information in complex, variable, and anatomically dispersed neural networks.
• Existing statistical methods for functional brain imaging have limitations in detecting these distributed representations.

Purpose of the Study:

• To identify limitations of current functional brain imaging statistical methods.
• To introduce and validate a novel whole-brain multivariate approach, sparse-overlapping-sets (SOS) LASSO, for detecting distributed neural information.
• To re-evaluate the anatomical distribution of neural representations, particularly for face perception.

Main Methods:

• Simulations using neural network models to test contemporary statistical methods (univariate contrast, searchlight, L1/L2 regularization) and the proposed SOS LASSO.
• Application of SOS LASSO to functional magnetic resonance imaging (fMRI) data to discriminate between visually presented faces and other stimuli.
• Further application to discriminate visual scenes from other stimuli.

Main Results:

• Contemporary methods exhibit critical blind spots in detecting distributed neural information.
• SOS LASSO successfully identifies network-distributed information, overcoming limitations of existing approaches in simulations.
• Applied to fMRI data, SOS LASSO reveals a widely distributed network for face perception across all four lobes, contrasting with localized findings from other methods.
• SOS LASSO also detects localized signals for visual scene discrimination, demonstrating its versatility.
• Decoding accuracy remains high even when canonical face/place systems are removed, confirming the validity of the distributed signal.

Conclusions:

• Structured sparsity, as implemented in SOS LASSO, offers an unbiased method for assessing functional localization claims.
• Prior functional imaging studies may have underestimated the distributed nature of neurocognitive representations.
• The findings suggest that neural representations, especially for complex stimuli like faces, are more widely distributed across the brain than previously assumed.