Neural responses to binocular in-phase and anti-phase stimuli

This study utilized SSVEPs to demonstrate that a two-stage contrast gain-control model incorporating parallel monocular channels, but not necessarily phase selectivity, effectively explains neural responses to binocular in-phase and anti-phase stimuli.

The Gist

Imagine your brain as a highly sophisticated sound mixing board, and your two eyes as two separate microphones picking up the world. Usually, when both microphones hear the same song, the brain blends them into one rich, clear stereo track. But what happens when the microphones pick up slightly different songs, or even the same song played backwards? Does the brain mash them together, cancel them out, or keep them separate?

This paper by Bruno Richard and Daniel Baker is like a detective story where they try to figure out exactly how the brain's "mixing board" works when the two eyes get conflicting signals.

The Experiment: Flickering Lights as a Test

Instead of showing people complex pictures, the researchers used a simple trick: they flashed a striped pattern (like a zebra crossing) on a screen at a steady rhythm (3 times a second).

• The "On/Off" Flicker: Imagine a light turning on and off. This creates a distinct "pulse" in the brain's electrical activity.
• The "Counterphase" Flicker: Imagine the stripes on the screen flipping colors instantly (black becomes white, white becomes black). This creates a different kind of electrical pulse.

They showed these lights to one eye, both eyes, or both eyes with the patterns shifted so they were "out of sync" (like two people clapping but one is always clapping when the other is opening their hands).

They measured the brain's response using EEG (a cap with sensors on the head), looking for a specific electrical "hum" that matched the flickering light.

The Big Question: How Does the Brain Mix the Signals?

For decades, scientists have had a theory called the "Two-Stage Gain Control Model." Think of this model as a recipe for how the brain mixes visual signals:

1. Stage 1 (The Pre-Processor): Each eye processes its own image first, adjusting for brightness and contrast (like a volume knob).
2. Stage 2 (The Mixer): The brain combines the two eyes' signals.
3. The Twist: The theory suggests there might be "parallel channels." Imagine a main highway where the two eyes merge into one big lane (Binocular), but there are also small side roads where each eye keeps driving separately (Monocular) all the way to the finish line.

The Discovery: The Brain Keeps the Side Roads Open

The researchers tested this by playing with the timing of the lights.

The Surprise: When they showed the eyes lights that were perfectly out of sync (temporal anti-phase), a purely "mixing" brain model predicted the signals would cancel each other out completely, leaving silence.

But the brain didn't go silent. It still hummed at the original frequency.

The Analogy: Imagine two people shouting the same word at the exact same time, but one says it normally and the other says it backwards. If they were perfectly mixed, you'd hear nothing. But in this experiment, the brain was still hearing the individual voices.

This proved that the "side roads" (the parallel monocular channels) are real. Even when the brain is trying to combine the eyes, it keeps a separate line open for each eye's individual signal. If the main mix gets messy or cancels out, the brain can still "hear" the individual inputs from the side roads.

What Didn't Matter: The "Phase" Confusion

The researchers also wondered if the spatial arrangement of the stripes (whether the black lines in the left eye lined up perfectly with the black lines in the right eye) changed the mix.

In other experiments (like judging which image looks brighter), this alignment matters a lot. But in this brain-scan experiment, it didn't matter at all. Whether the stripes were aligned or shifted, the brain's electrical hum was the same.

The Metaphor: It's like listening to a choir. If the singers are slightly out of tune with each other (spatial phase), it might sound messy to your ears (psychology), but the energy of the choir (the brain's electrical hum) remains the same. The brain's "volume meter" doesn't care about the tiny misalignments; it just cares about the overall noise.

The Conclusion: A Flexible, Robust System

The paper concludes that the "Two-Stage Model" is still the best way to describe how we see, but with one crucial update: The brain never fully shuts off the individual eyes.

Even when we are fusing two images into one, the brain maintains a backup track for each eye. This is a safety mechanism. If the two eyes disagree too much (like in a visual illusion or rivalry), the brain can fall back on the individual signals rather than getting confused by a failed mix.

In short: Your brain is a master mixer that loves to blend your two eyes into one perfect picture, but it always keeps the individual microphones live, just in case the mix goes wrong.

Technical Summary

1. Problem Statement

The human visual system integrates inputs from both eyes to form a unified percept. While compatible inputs lead to fusion, incompatible inputs (e.g., differing phases) can cause rivalry or diplopia. Computational models, specifically the two-stage contrast gain-control model (Meese et al., 2006; Georgeson et al., 2016), have successfully explained psychophysical data regarding binocular summation and discrimination. These models propose mechanisms including:

• Non-linear transduction (rectification).
• Interocular suppression (gain control).
• Parallel monocular channels (preserving signals from individual eyes).
• Phase-selective channels (responding to specific spatial phases).

However, previous neuroimaging applications of this model were limited to conditions where stimuli presented to both eyes had identical phases. It remained unclear whether this architecture could explain neural responses to incompatible inputs (specifically anti-phase stimuli) and whether monocular channels persist at the neural level during binocular viewing. This study aims to bridge the gap between psychophysical models and neural data by testing these models against Steady-State Visually Evoked Potentials (SSVEPs) under varying spatial and temporal phase relationships.

2. Methodology

Participants and Apparatus

• Subjects: 15 observers with normal or corrected-to-normal vision.
• Stimuli: A horizontal sinusoidal grating (3 cycles/degree) presented at 3 Hz.
• Display: Gamma-corrected ViewPixx 3D display (120 Hz refresh) with Nvidia stereo shutter goggles to ensure minimal crosstalk.
• EEG: Recorded from 64 electrodes (10/20 system) with a focus on occipital sites (Oz, POz, O1, O2).

Experimental Design
The study utilized two flicker protocols and varied the phase relationships between the left and right eyes:

1. On/Off Flicker: Stimulus alternates between 0% and 100% contrast. This generates responses at the fundamental frequency (3 Hz) and integer harmonics.
2. Counterphase Flicker: Stimulus alternates phase (black becomes white). This generates responses primarily at even harmonics (6 Hz, 12 Hz, etc.) due to frequency doubling.

Conditions:

• Monocular: One eye stimulated, the other gray.
• Binocular In-Phase: Both eyes receive identical stimuli (Spatial $\Delta\phi=0$, Temporal $\Delta\phi=0$).
• Binocular Anti-Phase:
  • Spatial Anti-Phase: Gratings shifted 180° relative to each other.
  • Temporal Anti-Phase: One eye's contrast increases while the other decreases (On/Off only).
  • Combined Anti-Phase: Both spatial and temporal anti-phase.

Analysis

• SSVEP amplitudes were calculated via Fourier transform of a 10-second window.
• Signal-to-Noise Ratios (SNRs) were computed at fundamental and harmonic frequencies.
• Statistical significance was assessed using permutation tests on median SNRs.

Modeling Approach
The authors progressively tested computational models against the observed SNR data:

1. Linear Sum: Simple addition of inputs (expected to fail).
2. Linear Sum with Rectification: Added half-wave rectification.
3. Two-Stage Gain Control (No Monocular): Standard model with interocular suppression but no parallel monocular output.
4. Two-Stage with Parallel Monocular Channels: Adds channels that preserve monocular signals throughout the architecture (Georgeson et al., 2016).
5. Full Model: Includes phase-selective channels (positive/negative phase) in addition to parallel monocular channels.

Models were fitted by minimizing the sum of squared errors between model outputs and observer median SNRs.

3. Key Results

Neural Response Patterns

• On/Off Flicker: Generated responses at 3 Hz (fundamental) and all integer harmonics.
• Counterphase Flicker: Generated responses primarily at even harmonics (6 Hz, 12 Hz, 18 Hz), with negligible fundamental response.
• Phase Effects:
  • Temporal Anti-Phase (On/Off): Significantly reduced the amplitude of the fundamental frequency (3 Hz) compared to in-phase conditions, but did not abolish it. Crucially, odd harmonics (9 Hz, 15 Hz) were still present.
  • Spatial Phase: No statistically significant differences in SNR were found between spatial in-phase and anti-phase conditions for either flicker type.

Model Performance

• Linear Models: Failed to predict harmonic responses and predicted zero response for temporal anti-phase On/Off stimuli (due to cancellation).
• Two-Stage (Binocular Only): Could generate responses for anti-phase stimuli but only at even harmonics (6 Hz, 12 Hz). It failed to explain the presence of the 3 Hz fundamental and odd harmonics observed in the data.
• Two-Stage with Parallel Monocular Channels: Successfully captured the data across all conditions. By preserving monocular signals, the model could generate the fundamental (3 Hz) and odd harmonics even when binocular inputs were in temporal anti-phase.
• Phase-Selectivity: Adding phase-selective channels did not improve model fit ($R^2$ remained unchanged), suggesting spatial phase does not significantly modulate SSVEP amplitudes at high contrast in this context.

4. Key Contributions

1. Validation of Monocular Channels in Binocular Vision: The study provides direct neural evidence (via SSVEP) that monocular signals persist during binocular viewing, even when stimuli are temporally anti-phase. This supports the "parallel monocular channel" hypothesis over models where only a summed binocular signal reaches higher processing stages.
2. Differentiation of Flicker Protocols: The research demonstrates that On/Off flicker is critical for detecting monocular contributions in binocular anti-phase conditions. While counterphase flicker suppresses the fundamental frequency (masking monocular signals), On/Off flicker reveals them via the persistence of the 3 Hz response.
3. Refinement of the Gain-Control Model: The authors confirm that the two-stage contrast gain-control model remains the most robust framework for binocular combination, provided it includes parallel monocular channels. They demonstrate that phase-selectivity, while important for psychophysics, may not be necessary to explain SSVEP amplitudes under high-contrast conditions.
4. Ocularity Invariance: The results reinforce the concept of ocularity invariance at high contrasts, where binocular and monocular responses are of similar magnitude, contrasting with the sensitivity gains seen at low contrasts.

5. Significance

This work is significant because it successfully bridges psychophysical theory and neurophysiological data.

• Theoretical Impact: It resolves a debate regarding whether monocular signals are discarded after binocular summation. The data confirms they are preserved and contribute to the final neural response, necessitating their inclusion in computational models of early vision.
• Methodological Impact: It highlights the utility of SSVEPs with specific temporal phase manipulations (temporal anti-phase) as a tool to isolate monocular vs. binocular processing streams.
• Clinical/Practical Impact: The robustness of the two-stage gain-control model across different modalities (behavioral and neural) suggests it is a reliable framework for understanding visual processing disorders or designing visual displays that rely on binocular integration.

In conclusion, the paper establishes that a two-stage contrast gain-control model with parallel monocular channels is the minimal necessary architecture to explain neural responses to complex binocular stimuli, while spatial phase selectivity appears less critical for SSVEP amplitude generation at high contrasts.