Distributed representational encoding of food attributes in ventral visual cortex

Using fMRI and representational similarity analysis, this study demonstrates that the ventral visual cortex encodes food attributes through a functional dissociation where the lateral occipitotemporal cortex represents both visual and subjective properties, while the fusiform gyrus selectively encodes perceived caloric content beyond visual similarity.

The Gist

Imagine your brain is a massive, high-tech kitchen. When you look at a picture of a juicy burger or a crisp apple, your brain doesn't just see "food." It instantly runs a complex analysis: What does this look like? Does it look tasty? Is it healthy? How many calories are in it?

For a long time, scientists thought the part of the brain responsible for seeing objects (the Ventral Visual Cortex) was like a simple camera lens. They believed it only processed the "pixels"—the colors, shapes, and textures. They thought the "judgment" part (deciding if it's healthy or high-calorie) happened later, in a different part of the brain like the "manager's office" (the frontal lobes).

This study, however, suggests that the "camera lens" is actually much smarter than we thought. It's not just taking a photo; it's already starting to write the review.

Here is the breakdown of what the researchers found, using some simple analogies:

The Setup: The "Food Tasting" Experiment

The researchers put 25 women in an MRI machine (a giant camera that takes pictures of brain activity). They showed them 96 different pictures of food, from donuts to salads. While the participants looked at the pictures, they had to do a simple task (pressing a button when a colored cross appeared) so their brains would focus on the food images without them having to consciously think about the food.

After the scan, the participants rated every single food picture on a scale of 1 to 100 for:

• How tasty it looked.
• How healthy it looked.
• How many calories they thought it had.

The Detective Work: RSA (The "Similarity Map")

Instead of just looking at which brain area "lit up" the brightest (like a simple on/off switch), the researchers used a technique called Representational Similarity Analysis (RSA).

Think of RSA like a social network for brain cells.

• If your brain cells react to a donut and a cake in almost the same way, the brain puts them in the same "friend group."
• If your brain cells react to a donut and a broccoli very differently, they are put in different groups.

The researchers asked: Do the brain's "friend groups" match up with how the humans rated the food?

The Big Discovery: Two Different "Chefs" in the Kitchen

The study found that the visual part of the brain isn't one single room; it has two distinct "chefs" (regions) that handle food information differently.

1. The "Generalist" Chef: LOTC (Lateral Occipitotemporal Cortex)

• Location: A bit further back in the visual processing line.
• What it does: This region is like a well-traveled food critic. It looks at the food and says, "This looks like a high-calorie treat, but it also looks unhealthy."
• The Finding: The brain activity here matched the participants' ratings for both "How many calories?" and "How healthy?"
• The Analogy: This region is like a general manager who looks at the menu and considers the whole picture: taste, health, and energy. It blends the visual look with the nutritional idea.

2. The "Specialist" Chef: Fusiform Gyrus

• Location: A bit further forward in the visual line (often associated with face recognition, but here it's the "food specialist").
• What it does: This region is like a nutritionist who only cares about the calorie count.
• The Finding: This area was very specific. It organized the food images based only on how many calories the participants thought they had. It didn't care much about whether the food looked "healthy" or "tasty."
• The Analogy: Imagine a robot that scans a burger and instantly calculates its energy output, ignoring the fact that it looks delicious or that it's bad for your heart. It isolates the "calorie" dimension perfectly, even after the researchers mathematically removed the visual similarities (like color or shape) from the data.

The Surprise: The "Manager's Office" was Quiet

The researchers also looked at the Orbitofrontal Cortex (OFC) and Insula. In previous studies, these are the "Manager's Office" and the "Body Sensor" where we usually expect to find feelings of hunger, reward, and value.

• The Expectation: They thought these areas would show complex patterns matching the "tasty" or "healthy" ratings.
• The Reality: These areas lit up (showed activity) when looking at high-calorie vs. low-calorie food, but they didn't show the complex patterns the researchers were looking for.
• Why? The researchers suggest this is because these areas are like volatile weather systems. They change based on how hungry you are right now, your mood, or what you just ate. Because the study controlled for hunger (everyone ate a snack 2 hours before), the "weather" was too calm to see the complex patterns. The brain was in "neutral" mode, so the deep value judgments weren't fully active.

The Takeaway: Your Eyes Are Already Judging

The most exciting part of this paper is the conclusion: Your visual system isn't just a camera.

When you look at a picture of a greasy pizza, your brain's visual cortex isn't just seeing "red sauce and white cheese." It is already organizing that image in a way that reflects your knowledge that "this is high-calorie."

• The "Generalist" area (LOTC) sees the whole story: "This is tasty but unhealthy."
• The "Specialist" area (Fusiform) zooms in on the energy: "This is pure fuel."

It turns out that the part of your brain that sees the food is deeply connected to the part of your brain that values the food. They aren't separate steps; they are happening together, right at the moment you look at the food.

In short: You don't just "see" food and then "decide" if it's good for you. Your brain's visual system has already started the nutritional analysis before you even realize you're hungry.

Technical Summary

1. Problem Statement

While the neural processing of visual food features is well-documented, it remains unclear how subjective properties (e.g., palatability, perceived health) and nutritional attributes (e.g., caloric content) are encoded within the neural representational structure of the brain. Previous research has largely relied on univariate analyses (comparing mean activation levels), which often yield inconsistent results and fail to capture the distributed patterns of activity that may encode complex food information. The study aims to determine:

• Whether subjective ratings and objective nutritional properties are reflected in multivariate brain activity patterns.
• How visual features, subjective judgments, and higher-order food properties are differentially represented across the cortical hierarchy, specifically within the ventral visual stream.

2. Methodology

Participants and Design

• Sample: 25 female participants (selected to reduce variability in neural responses to food cues).
• Procedure: Two scanning sessions one week apart. Participants viewed 96 food images (balanced for high/low calories and sweet/savory) during fMRI.
• Task: An orthogonal color-discrimination task (identifying the color of a fixation cross) was used to maintain attention without explicitly engaging food valuation during scanning.
• Post-Scan Ratings: Participants rated each stimulus on four dimensions: palatability, familiarity, perceived health value, and perceived caloric content.

Stimuli and Imaging

• Stimuli: 96 food images standardized for resolution and background.
• MRI: 3T Siemens Prisma Fit scanner. High-resolution T1 and multiband EPI sequences (TR=1500ms).
• Preprocessing: Performed using fMRIPrep (25.2.2), including motion correction, susceptibility distortion correction, and normalization to MNI152 space.

Analytical Approach

• Univariate Analysis: A General Linear Model (GLM) contrasted high-calorie vs. low-calorie foods to identify regions with mean activation differences.
• Representational Similarity Analysis (RSA):
  • Neural Data: Stimulus-specific beta estimates were extracted from 6 Regions of Interest (ROIs): V1, Lateral Occipitotemporal Cortex (LOTC), Fusiform Cortex, Orbitofrontal Cortex (OFC), Insula, and Dorsolateral Prefrontal Cortex (DLPFC).
  • Representational Dissimilarity Matrices (RDMs): Neural RDMs were computed using correlation distance between multivoxel patterns.
  • Model RDMs: Several computational and behavioral models were constructed to test against neural data:
    • Visual Models: Gabor filters (low-level), CORnet-S (V4 and IT layers for mid/high-level), and Color histograms.
    • Subjective Models: Behavioral ratings for palatability, health, and perceived calories.
    • Categorical Models: Binary distinctions for high/low calories and sweet/savory.
  • Control Analyses: Partial correlations were used to determine if effects persisted after controlling for visual similarity (e.g., testing if "Perceived Calories" explained variance beyond "CORnet IT").

3. Key Contributions

• Integration of Multivariate and Univariate Methods: The study bridges the gap between traditional activation mapping and modern representational geometry, showing how they complement each other in food perception research.
• Dissociation within the Ventral Stream: It provides evidence for a functional dissociation between the LOTC and the Fusiform Gyrus, two adjacent regions often grouped together in object recognition.
• Beyond Visual Statistics: The study demonstrates that food representations in the visual cortex encode information that cannot be explained by low-level visual features (color, texture) or even high-level object recognition models (CORnet), suggesting the visual system integrates nutritional and subjective knowledge.

4. Key Results

Univariate Findings

• Significant activation differences between high- and low-calorie foods were found in V1, LOTC, Fusiform Gyrus, OFC, and Insula. This confirms that mean activation levels are sensitive to caloric content.

Multivariate (RSA) Findings

• LOTC (Lateral Occipitotemporal Cortex):
  • Showed systematic correspondence with high-level visual models (CORnet V4/IT).
  • Also correlated with subjective dimensions (perceived caloric content and health value).
  • Control Analysis: The calorie effect remained significant after controlling for visual models but was reduced when controlling for health ratings, suggesting LOTC encodes a shared axis of energy density and health appraisal.
• Fusiform Cortex:
  • Exhibited a selective association with perceived caloric content.
  • Unlike LOTC, it did not show a significant correlation with perceived health value.
  • Control Analysis: The calorie effect persisted even after controlling for visual hierarchy (CORnet), low-level image properties, and health ratings. This indicates a specific encoding of caloric information independent of visual similarity.
• V1: Dominated by low-level visual models (Gabor), as expected.
• OFC, Insula, DLPFC:
  • Despite showing univariate activation differences for high vs. low calories, no significant RSA effects were found for subjective models in these regions.
  • Interpretation: These regions may encode value in a state-dependent or categorical manner (mean activation shifts) rather than organizing fine-grained stimulus representations based on specific attributes like calories or health.

5. Significance and Conclusion

The study challenges the view of the ventral visual stream as a purely perceptual hierarchy. Instead, it demonstrates that subjective and nutritional dimensions are embedded within visual representational spaces at multiple levels of abstraction:

1. LOTC acts as an integrative hub where visual object features are structured by abstract, behaviorally relevant dimensions (both calorie and health).
2. Fusiform Cortex appears to selectively encode visual patterns predictive of caloric content, likely reflecting learned statistical regularities linking visual appearance to energy density.

These findings suggest that the brain's visual system does not merely process "what" an object looks like, but also "what it is worth" or "what it contains," integrating sensory input with nutritional and subjective knowledge. This has implications for understanding dietary choices and the neural mechanisms underlying food-related disorders. The lack of RSA effects in valuation regions (OFC/Insula) highlights the complexity of value coding, which may be more dynamic and context-dependent than the stable stimulus-locked representations found in the visual cortex.