Defining a functional hierarchy of millisecond time: from visual stimulus processing to duration perception

Using ultra-high field 7T fMRI and neuronal modeling, this study reveals a functional hierarchy in millisecond time processing where parietal and premotor cortices encode precise, clustered duration maps for task readout, while the rostral SMA, inferior frontal cortex, and anterior insula generate categorical, mean-centered representations that correlate with subjective perceptual boundaries.

The Gist

Imagine your brain is a massive, high-tech factory dedicated to processing time. For a long time, scientists knew this factory existed and that many different rooms (brain areas) were involved in timing. But they didn't quite understand the assembly line: how does a raw signal of "time passing" get turned into your conscious feeling of "that was a long time" or "that was short"?

This study, using a super-powerful 7-Tesla MRI scanner (think of it as a microscope for the brain), mapped out exactly how this factory works. Here is the story of how your brain processes milliseconds, explained through a few simple analogies.

1. The Raw Material: The "Accumulating Rain"

Location: The back of your brain (Visual Cortex).

When you see a light flash for a split second, the very first thing your brain does is act like a bucket catching rain. The longer the light stays on, the more "water" (neural activity) fills the bucket.

• What they found: In the early visual areas, the brain doesn't have a specific "timer" for 0.3 seconds or 0.7 seconds. Instead, it just counts up. The longer the stimulus, the stronger the signal. It's a simple, monotonic "ramp" of activity.
• The Analogy: Think of a water hose filling a bucket. You don't need to know exactly how many gallons are in there yet; you just know the longer the hose runs, the fuller the bucket gets. This is the "encoding" stage.

2. The Sorting Facility: The "Library of Time"

Location: The middle of your brain (Parietal and Motor areas).

Once the water is in the bucket, the signal moves to the middle of the brain. Here, the "filling up" stops, and the brain starts organizing.

• What they found: In these areas, the brain creates specific "shelves" or "maps." There are neurons that love 0.2 seconds, others that love 0.4 seconds, and others that love 0.8 seconds. They are arranged neatly next to each other on the brain's surface, like books on a shelf.
• The Analogy: Imagine a library. Instead of a giant pile of books (the raw signal), the brain now has a librarian who sorts them into specific sections. One section is for "Short Stories," another for "Novels," and another for "Encyclopedias." This is the "readout" stage, where the brain identifies exactly what duration it just saw.

3. The Decision Room: The "Judge's Gavel"

Location: The front of your brain (Frontal Cortex, Insula, and top of the motor area).

Finally, the signal reaches the front of the brain, where the real thinking happens. This is where you decide: "Was that longer or shorter than the 0.5-second reference I was thinking of?"

• What they found: Here, the brain stops caring about every tiny fraction of a second. Instead, it focuses on the middle point (the average). The neurons here act like a judge sitting at a gavel. They don't care if the time was 0.49s or 0.51s; they care about the boundary.
• The Analogy: Imagine a courtroom judge. The judge doesn't need to know the exact weight of every grain of sand on the scale. They just need to know: "Is this pile heavier or lighter than the standard weight?" These neurons represent the "cutoff line" in your mind. Interestingly, the study found that the position of this "gavel" in your brain matches your personal perception. If you tend to think things are shorter than they are, your brain's "gavel" is set slightly differently than someone else's.

The Big Picture: A Functional Hierarchy

The study reveals that time isn't processed in one big lump. It flows through a hierarchy:

1. Input (Back of brain): "Something is happening, and it's getting longer." (The Rain Bucket)
2. Processing (Middle of brain): "Okay, that specific duration was 0.6 seconds." (The Library)
3. Decision (Front of brain): "Is 0.6 seconds longer than my reference of 0.5? Yes. I will press the button." (The Judge)

Why This Matters

This research is like finding the blueprint for a clock. It shows us that our subjective experience of time isn't magic; it's built step-by-step by different teams of neurons doing specific jobs.

• The "Why": It explains why we can sometimes be fooled by time (e.g., time flying when we're having fun). If the "Judge" in the front of the brain changes its mind about where the boundary is, our whole perception of time shifts.
• The "How": It proves that the brain transforms a simple physical event (a light flashing) into a complex, subjective feeling (my perception of time) by moving it through a specialized assembly line.

In short, your brain doesn't just "feel" time; it builds it, layer by layer, from a simple sensory signal into a complex decision.

Technical Summary

1. Problem Statement

While the neural network involved in subsecond time processing (the "timing network") is well-documented, the specific functional roles of unimodal neuronal tuning (where neurons respond maximally to a specific duration) across the cortical hierarchy remain unclear. Previous studies have identified "chronomaps" (topographically organized duration preferences) in various regions, but it is unknown:

• How duration tuning properties change as information flows from early sensory areas to higher-order frontal regions.
• Whether these changes reflect distinct functional stages (e.g., encoding vs. decision-making).
• How these neural tuning properties directly relate to subjective duration perception and behavioral decision boundaries.

2. Methodology

Participants and Data Source

• Subjects: 13 healthy human participants.
• Data Source: The study utilized existing 7T fMRI data from a previous experiment (Centanino et al., 2023/2024), acquired at the Spinoza Centre for Neuroimaging.

Experimental Paradigm

• Task: A visual duration discrimination task. Participants judged whether a visual stimulus (a colored Gaussian noise patch) was longer or shorter than an internalized reference duration of 0.5 s.
• Stimuli: Comparison durations ranged from 0.2 s to 0.8 s (in 0.1s steps).
• Design: Stimuli were presented at four spatial locations (0.9° and 2.5° eccentricity in lower-left/right quadrants) to dissociate temporal processing from spatial attention.
• Acquisition: 7T fMRI with high spatial resolution (1.8 mm isotropic voxels) and temporal resolution (TR = 1.32 s).

Analytical Pipeline

1. GLM Analysis: A General Linear Model (GLM) was fitted to the fMRI data, modeling the BOLD response at stimulus offset for all 24 combinations of duration and position.
2. Population Receptive Field (pRF) Modeling: Instead of standard univariate analysis, the authors applied a duration-tuned pRF model.
   • The model assumes a Gaussian response function to duration: $nr \sim e^{-(d - \mu_d)^2 / 2\sigma_d^2}$.
   • Parameters: $\mu_d$ (duration preference) and $\sigma_d$ (sensitivity). The model is invariant to spatial position.
   • This allowed the extraction of a preferred duration map for every vertex on the cortical surface.
3. Region of Interest (ROI) Definition: ROIs were defined using high-resolution parcellation atlases (HCP MMP 1.0 and Sereno et al.) based on group-level activation clusters. 93 ROIs were grouped into 9 functional streams: Ventral Visual (VV), Lateral Visual (LV), Intraparietal Sulcus (IPS), Inferior Parietal (IP), Motor-Somatosensory (mot-som), Supplementary Motor Area (SMA), Premotor (PM), Anterior Insula (AI), and Inferior Frontal (IF).
4. Statistical & Spatial Analysis:
   • Linear Mixed Effects (LME) Models: To test for differences in median $\mu_d$ across streams and ROIs.
   • Categorization: Durations were binned into five categories (short to long) to analyze distribution shifts.
   • Spatial Statistics: Local and Global Moran's I statistics and Variograms (nugget and range) were computed to assess the topographic clustering and spatial autocorrelation of duration preferences.
   • Correlation Analysis: Kendall correlations were computed between neural preferences and behavioral Point of Subjective Equality (PSE) to link tuning to perception.

3. Key Contributions

• Functional Hierarchy Definition: The study establishes a three-stage functional hierarchy for visual time processing: Encoding (monotonic accumulation), Readout (unimodal mapping), and Categorization (subjective boundary representation).
• Integration of Tuning and Topography: It demonstrates that not only do preferred durations change along the hierarchy, but the spatial organization (clustering) of these preferences also evolves, becoming more structured in higher-order areas.
• Neural Basis of Subjective Time: It provides direct evidence linking specific neuronal populations (in rostral SMA and anterior insula) to the subjective decision boundary (PSE) used by participants, suggesting these areas encode the "criterion" for time perception.
• SMA Segregation: It reveals a rostro-caudal functional gradient within the SMA, distinguishing between caudal regions (duration readout) and rostral regions (categorical boundary representation).

4. Key Results

A. Changes in Duration Preferences ($\mu_d$)

• Occipital Visual Areas (VV & LV): Showed a preference for longer durations (skewed toward 0.6–0.8 s). This suggests a monotonic ramping response (accumulation of sensory evidence) rather than discrete tuning.
• Parietal, Premotor, and Caudal SMA: Showed a broad distribution of preferences covering the full range of presented durations (0.2–0.8 s). This indicates a population of unimodally tuned neurons capable of reading out specific durations.
• Rostral SMA, Anterior Insula (AI), and Inferior Frontal (IF): Showed preferences centered around the mean duration (0.5 s). This suggests a shift toward categorical representation, where neurons encode the boundary used for decision-making.
• Motor-Somatosensory Areas: Showed a preference for short durations, likely reflecting motor preparation rather than pure duration tuning.

B. Spatial Organization (Topography)

• Clustering: Occipital areas showed weaker spatial autocorrelation (lower Moran's I, higher nugget), consistent with monotonic accumulation.
• Structured Maps: Parietal, premotor, and frontal areas showed strong spatial clustering (high Moran's I, lower nugget), indicating organized "chronomaps" where similar duration preferences are spatially contiguous.
• Local Moran's I: In areas with broad tuning (parietal/SMA), both "High-High" and "Low-Low" associations coexisted. In areas with narrow tuning (AI/IF), specific associations dominated, reflecting the dominance of the mean duration.

C. Long-Range Relationships

• Hierarchical Clustering: Correlation analysis of median preferences across streams revealed three distinct functional modules:
  1. Visual Cluster: VV and LV (Encoding).
  2. Cognitive Cluster: IPS, AI, and IF (Integration and Categorization).
  3. Motor Cluster: IP, SMA, PM, and mot-som (Readout and Action).
• Negative Correlations: Strong negative correlations were found between the visual cluster and the motor/cognitive clusters, suggesting a transformation of the signal from sensory accumulation to abstract representation.

D. Link to Perception

• PSE Correlation: There was a significant positive correlation between the Point of Subjective Equality (PSE) and the median duration preference in the Anterior Insula (bilateral) and Rostral SMA (specifically area a32pr).
• Gradient: The strength of the correlation between neural preference and PSE increased progressively from occipital to frontal regions, confirming that the link to subjective perception is built up along the cortical hierarchy.

5. Significance

This paper provides a comprehensive mechanistic framework for how the brain transforms raw temporal information into a subjective percept.

1. Mechanism of Perception: It moves beyond simply identifying "timing areas" to explaining how they work. It posits that time perception is not a single process but a cascade: sensory accumulation $\rightarrow$ topographic readout $\rightarrow$ categorical boundary formation.
2. Subjectivity: By linking the activity in the anterior insula and rostral SMA to individual PSEs, the study suggests that the subjective experience of time is rooted in the specific tuning properties of "boundary neurons" that integrate internal states with external sensory input.
3. Methodological Advance: The application of pRF modeling to duration (rather than just spatial or visual features) at 7T resolution sets a new standard for investigating temporal processing in the human brain, allowing for the mapping of functional gradients with high anatomical precision.
4. Clinical/Computational Implications: Understanding the specific breakdown of these stages could inform research into timing deficits in neurological disorders (e.g., Parkinson's, Schizophrenia) and improve computational models of temporal decision-making.