Visual representations are dominated by intrinsic fluctuations correlated between areas

Area of Science:

• Neuroscience
• Cognitive Science
• Neuroimaging

Background:

• Trial-to-trial variability in visually evoked responses is linked to intrinsic cortical dynamics.
• The role of these dynamics in sensory processing and neural representation remains a key challenge.

Purpose of the Study:

• To investigate how intrinsic cortical dynamics influence representational geometry in human visual areas.
• To determine if intrinsic dynamics exaggerate representational similarity between brain regions.
• To characterize stimulus-driven representations by disentangling them from intrinsic dynamics.

Main Methods:

• Functional magnetic resonance imaging (fMRI) was used to measure brain responses to natural images.
• Representational Dissimilarity Matrices (RDMs) were constructed from fMRI response patterns.
• Analyses compared RDMs derived from the same trials (shared dynamics) versus separate trials (stimulus-driven).

Main Results:

• Representational geometries were highly similar across visual areas when using the same trials, but distinct when using separate trials.
• Coherent fluctuations in regional-mean activation, reflecting intrinsic dynamics, explained the exaggerated similarity.
• Stimulus-driven representations showed clearer distinctions: Gabor model for V1-V3, animate-inanimate model for lateral occipital cortex.

Conclusions:

• Intrinsic cortical dynamics significantly impact and can obscure the true stimulus-driven representational geometry.
• Disentangling intrinsic dynamics is crucial for accurately characterizing neural representations.
• Different visual areas exhibit distinct representational strategies, from low-level features to high-level semantics.