Improving Understanding of Visual Snow by Quantifying its Appearance and Effect on Vision

Abstract

Purpose: Visual snow is the hallmark of the neurological condition visual snow syndrome (VSS) but the characteristics of the visual snow percept remain poorly defined. This study aimed to quantify its appearance, interobserver variability, and effect on measured visual performance and self-reported visual quality.

Methods: Twenty-three participants with VSS estimated their visual snow dot size, separation, luminance, and flicker rate by matching to a simulation. To assess whether visual snow masks vision, we compared pattern discrimination thresholds for textures that were similar in spatial scale to visual snow as well as more coarse than visual snow, in participants with VSS, and with and without external noise simulating visual snow in 23 controls.

Results: Mean and 95% confidence intervals for visual snow appearance were: size (6.0, 5.8–6.3 arcseconds), separation (2.0, 1.7–2.3 arcmin), luminance (72.4, 58.1–86.8 cd/m²), and flicker rate (25.8, 18.9–32.8 frames per image at 120 hertz [Hz]). Participants with finer dot spacing estimates also reported greater visibility of their visual snow (τ_b = −0.41, 95% confidence interval [CI] = −0.62 to −0.13, P = 0.01). In controls, adding simulated fine-scale visual snow to textures increased thresholds for fine but not coarse textures (F(1, 22) = 4.98, P = 0.036, η_p² = 0.19). In VSS, thresholds for fine and coarse textures were similar (t(22) = 0.54, P = 0.60), suggesting that inherent visual snow does not act like external noise in controls.

Conclusions: Our quantitative estimates of visual snow constrain its likely neural origins, may aid differential diagnosis, and inform future investigations of how it affects vision. Methods to quantify visual snow are needed for evaluation of potential treatments.

Visual snow syndrome (VSS) is a neurological condition with multiple visual symptoms¹ that impair quality of life.^2,3 Visual snow is the syndrome's defining symptom and often its most disturbing.⁴ While people consistently describe visual snow as tiny, dynamic dots across their vision,¹ rating scales suggest individual differences in appearance.^2,5 Visual snow also causes varying degrees of distraction and disruption, which are related to individual differences in self-rated visual snow appearance.^2,5,6 However, our understanding of the visual snow percept and its impact on daily visual experience is currently limited to qualitative descriptions. 

Additionally, it is unclear whether the visual snow percept impairs visual function. Although prior research has identified subtle abnormalities in visual perception in people with VSS,^7–9 how visual snow itself affects vision has not been directly assessed. A reasonable hypothesis is that, like external noise, visual snow may mask relevant information in the visual scene. According to self-reports, visual snow can mask vision,⁵ impair vision in dim lighting,¹ and affect reading and screen use.² Nevertheless, previous research shows that clinical measures of vision and tests of signal extraction from noise are typically normal in VSS.^7,8 However, one possibility is that standard test stimuli likely contain spatial details much coarser than visual snow and are therefore ill-suited to measuring potential noise-like interference. 

Therefore, in this study, we quantitatively measured the typical appearance and heterogeneity of visual snow for the following dimensions — dot size, density, brightness, and speed — via a psychophysical matching task to simulated visual snow. We then used our quantitative description of visual snow to contextualize its self-perceived severity, as assessed with a questionnaire. 

Last, we evaluated whether visual snow acts like external noise masking textures of a similar spatial scale but not coarser textures. In people without VSS, we confirmed that external noise simulating visual snow degrades fine, visual snow-like textures but not coarse textures. In people with VSS, we then inferred whether visual snow acts like external noise by comparing perception for fine (matched to each individual's visual snow) versus coarse textures. 

Methods

Participants

We measured visual snow appearance, perceived severity, and effect on texture perception in 23 individuals with VSS (mean age = 31, range = 21–45 years; 11 with migraine) and assessed the effect of simulated visual snow on texture perception in 23 controls (mean age = 28, range = 18–45 years). Previous research has found perceptual differences in VSS with a similar sample size.^7,9 

Participants were recruited from a database of previous participants and via advertisement within the University of Melbourne from July 2021 to June 2022. Participants were aged 18 to 45 years, as older age impairs performance on the texture perception task.^10,11 The relevant diagnostic criteria were used to classify participants with VSS¹ and migraine.¹² VSS onset within 12 months of hallucinogenic drug use was a basis for exclusion. Three VSS participants used neuroactive medication (fluvoxamine and gabapentin, venlafaxine, and fluoxetine). Controls were excluded if they reported more than four headaches a year, a history of headaches with features of migraine, or use of medications known to affect vision or cognition. 

A clinical eye examination was performed to ensure best corrected visual acuity of 6/7.5 or better, refractive error no more than ± 5.00 diopters (D) sphere and 2.00 D astigmatism, normal ocular health (pupil responses, ocular motility, slit lamp biomicroscopy examination, and fundus examination and photography) and 24-2 threshold visual fields (Compass Automated Fundus Perimeter, CenterVue SpA, Italy). Six prospective participants (3 controls and 3 participants with VSS) were excluded based on eye examination results. 

All protocols were approved by the University of Melbourne Human Research Ethics Committee and participants provided written informed consent in accordance with the Declaration of Helsinki. 

Questionnaire on the Perceived Severity of Visual Snow

A questionnaire was developed with input from an individual with VSS, requiring participants to categorize visual snow color¹³ and rate perceived severity (visibility, effort to ignore, annoyance, and visual interference) on an 11-point numerical scale in bright and dim lighting, and comment on impaired night vision (Supplementary Fig. S1). 

Computer-Based Vision Testing

Software was custom written in Python using PsychoPy3¹⁴ running on a Dell personal computer (Windows 10). Stimuli were displayed on a BenQ monitor (1920 × 1080 pixels, 53 × 30 cm, 120 Hz) calibrated using an OptiCal luminance meter (Cambridge Research Systems, Cambridge, UK). Participants viewed the monitor binocularly with refractive correction in dim lighting. 

Quantification of Visual Snow Appearance

We developed a matching task with input from an individual with VSS. Participants looked from one side of the screen to the other to compare their visual snow on the black side of the screen to an external stimulus mimicking visual snow (Fig. 1). The simulation was a 960 × 960-pixel image with a black background of 0.61 cd/m² upon which white dots were randomly placed with a minimum separation that controlled density. Matches were made in dim illumination, under which visual snow should be most noticeable, and for each dimension in isolation and not the visual snow percept as a whole. 

Figure 1.

 

Illustration of the matching task quantifying visual snow appearance. The participant's own visual snow, readily visible on the black left side of the screen, was compared to simulated visual snow on the right that consisted of randomly placed white dots on a black background. This illustration is not to scale, as in the experiment the image was viewed from a remote distance. However, this illustration is designed to be viewed at the intended size from a typical reading distance, as the simulation may appear non-uniform due to aliasing when minified.

Dot size was measured first. Pilot testing revealed that visual snow dots were too tiny to match by manipulating simulation dot size on the monitor. Therefore, participants viewed simulated visual snow (100 cd/m² one-pixel dots, minimum separation 60 pixels) through a mirror and walked backward until the simulation dots appeared equal in size to their visual snow. The maximum viewing distance was 11.5 meters (m). Participants then approached the mirror from the back of the room to make another match. Dot size was calculated as the visual angle subtended by one pixel at the average viewing distance. 

The remaining dimensions of visual snow appearance were measured at this individualized distance. Using a method of adjustment, participants manipulated the dimension of interest using the keyboard. 

To measure density, participants varied the minimum possible separation between 100 cd/m² dots in steps of 0.3 arcmin until the simulation was of comparable density to their visual snow. Minimum separation was taken as the average of four matches, two starting at a high density (small separation of 0.5 arcmin) and two at a low density (larger separation of 3.5 arcmin). 

To measure brightness, participants adjusted dot luminance in steps of 12 cd/m² from 100 cd/m² until simulation dots with a 2 arcmin separation appeared equal in brightness to their visual snow. The average of four measures was taken. Note that white dots on a black background can simulate most types of visual snow reported in the literature, which have been qualitatively categorized as “black and white (i.e., only black dots on white background, white dots on black background),” “clear,” or “always white.”¹³ While people with VSS more frequently report a single static type that is typically achromatic, colored dots or multiple types are not uncommon.¹³ Therefore, participants were instructed to base matches on their overall impression of visual snow dot brightness, regardless of color or the presence of multiple types. 

Flickering static was created by displaying multiple images of simulated visual snow (150 cd/m² dots, 2 arcmin separation) in sequence for the same number of frames. Flicker rate is inversely related to the number of frames per image. Participants adjusted the number of frames per image in steps of six frames until the simulation and visual snow flicker rate matched. Two measures starting from fast flicker (3 frames per image/40 images per second) and two from slow flicker were averaged (75 frames per image/1.6 images per second). 

Following measurement, participants rated how well they matched each dimension of their visual snow to the simulation on an 11-point numerical scale spanning responses from not at all (0) to exact match (10). 

Customized Test of Visual Interference

We evaluated whether visual snow masks vision in a similar fashion to external noise using dot textures customized to the measured spatial scale of visual snow. Texture stimuli were 960 × 960 pixel Glass patterns¹⁵ in which pairs of signal dots were arranged in concentric or radial forms (Figs. 2, 3). Replacing signal dots with noise consisting of randomly positioned dots degrades this global structure. Glass patterns are well-understood stimuli used to study how the visual system extracts global form signals from noise.^16,17 We measured the coherence (percentage of signal dots) required for discrimination between concentric and radial structure, similar to tasks used in other clinical populations^10,18 including VSS.⁷ However, this previous research used standard textures likely to be coarser in spatial scale than visual snow. 

Figure 2.

 

Illustration of fine textures for the customized test of visual interference. Participants discriminated between concentric and radial Glass patterns, shown sequentially in a randomized order. (A, B) Example fine textures without simulated visual snow, depicting a concentric (A) and radial (B) Glass pattern at 100% coherence on a uniform black background. (C) Example of a fine texture with external visual snow-like noise, consisting of a visual snow simulation superimposed on a 100% coherence concentric Glass pattern of the same spatial scale.

Figure 3.

 

Illustration of coarse textures for the customized test of visual interference. Participants discriminated between concentric and radial Glass patterns, shown sequentially in a randomized order. (A, B) Example coarse textures without simulated visual snow, depicting a concentric (A) and radial (B) Glass pattern at 100% coherence on a uniform black background. (C) Example of a coarse texture with external visual snow-like noise, consisting of a visual snow simulation superimposed on a 100% coherence concentric Glass pattern that was much coarser in spatial scale.

Here, we compared discrimination performance for novel Glass patterns with tiny, dense dots like visual snow (fine textures, 1 pixel dots; see Figs. 2A, 2B) and large, sparse dots (coarse textures, 8 pixel dots; see Figs. 3A, 3B). Fine textures contained eight times the total number of dots (signal plus noise) as coarse textures. For both texture types, dots were 80 cd/m² and dots in a signal pair were separated by 6 arcmin. 

Separate experiments were conducted for controls and those with VSS. In controls, we investigated whether external visual snow-like noise degraded fine textures more than coarse textures. Textures were presented with and without full screen simulated visual snow (6 arcseconds, 70 cd/m² dots with 2 arcmin minimum separation) superimposed that was representative of the average values measured in the VSS group. Fine textures (see Fig. 2) were matched in dot size and density to the simulated visual snow, consisting of 1362 dots (derived from simulation dot separation), and 6 arcseconds in size (one-pixel dots viewed from 964 cm). Coarse textures (see Fig. 3) contained dots that were eight times larger (48 arcseconds) and sparser (170 dots) than the simulated visual snow. 

In a separate experiment, each participant with VSS viewed unique coarse and fine textures through their own constant, internally generated visual snow percept without any superimposed external noise. Fine textures (see Figs. 2A, 2B) were individualized to each participant's estimated visual snow dot size and density from the matching task. To infer whether visual snow acted like external noise in controls, coarse textures (see Figs. 3A, 3B) were created using the same relative change in spatial scale, as dots were eight times wider and an eighth the number compared to fine textures. Note that, by design, this experiment in participants with VSS was not intended to enable a direct comparison with performance in controls, who viewed textures that differed in dot size and number. 

On each trial, a concentric and radial Glass pattern were each presented in separate 180 ms intervals in a randomized order with a 500 ms interstimulus interval. Participants chose which interval contained the concentric pattern. Seven coherence levels were presented with 30 repeats, across 3 runs that presented each level 10 times in a randomized order using a method of constant stimuli. Coherence levels were individually chosen to represent the psychometric function range as task performance varied between individuals. Fine and coarse textures were assessed in counterbalanced order. In controls, runs with and without simulated visual snow were interleaved. 

For each condition, the R package “quickpsy”¹⁹ was used to fit individual data with a psychometric function²⁰:  

\begin{eqnarray}{\rm{\Psi }}\!\left( x \right){\rm{\;}} = {\rm{\;\gamma }} + \left( {1 - {\rm{\gamma }} - {\rm{\lambda }}} \right)G\!\left( {x,\mu ,\sigma } \right)\end{eqnarray}

(1)

that described the sigmoidal change in proportion correct responses as a function of coherence level (x) using a cumulative Gaussian G(x, µ, σ) with mean µ and standard deviation σ, guess rate (γ) of 0.5 and a variable lapse rate (λ). 

The mean of the psychometric function was taken as the threshold coherence level (the strength of the form signal) for reliably discriminating between concentric and radial patterns. If visual snow degrades fine textures more than coarse textures, fine textures will require higher coherence for good performance. Alternatively, visual snow may increase task difficulty due to poorer discrimination precision, resulting in a shallower slope. The spread of the psychometric function, given by the standard deviation, is inversely related to the slope, such that higher values indicate a shallower slope. 

Analysis

Statistical analyses were conducted in SPSS Statistics 29 (IBM, Armonk, NY, USA). Intraclass correlation coefficients were calculated for each dimension of visual snow based on a single measurement, absolute-agreement, two-way mixed effects model. Pearson correlations were performed to determine whether estimates for the different dimensions were related. The relationship between visibility ratings and measured appearance was assessed by calculating Kendall's tau correlation coefficients for each dimension. To explore factors influencing the perceived severity of visual snow, we compared ratings in dim and bright lighting using Wilcoxon signed-rank tests and, for each lighting condition, performed Kendall's tau correlations between visibility ratings and ratings of perceived functional impact (interference with vision, effort to ignore, and annoyance). Correlations were uncorrected for multiple comparisons as this was an exploratory study. 

Thresholds and psychometric function spread were log transformed for analysis. Due to differences in the stimuli used for the experiment in controls and the experiment in participants with VSS, performance was analyzed for each group separately in the main analysis. For controls, separate 2-way repeated measures ANOVAs were performed for thresholds and spread, with texture type and presence of simulated visual snow as factors. For participants with VSS, thresholds and spread were analyzed using separate paired t-tests to compare performance for the individualized fine and coarse textures. 

Results

Typical Appearance and Heterogeneity of Visual Snow

Figure 4 shows the estimates for each dimension of visual snow. Participants consistently estimated that their visual snow contained tiny dots packed close together (see Supplementary Fig. S2 for a representative example). Estimates of dot luminance and flicker speed showed greater interindividual variation. Estimates of size, separation, luminance, and frames per image were unrelated (Pearson correlations, P > 0.5). All participants with VSS reported longstanding visual snow (earliest memory: n = 15, later onset: n = 8; duration: average = 23.4 years, and range = 3.4–45 years) that had either remained stable (n = 18) or worsened in the years since onset (n = 5). 

Figure 4.

 

Quantitative estimates of visual snow appearance. Individual data (small grey circles), mean (large black circles), and 95% confidence intervals (error bars) for estimated (A) dot size in seconds of arc, (B) minimum separation between dots in minutes of arc, which is inversely related to perceived visual snow density, (C) dot luminance (cd/m²) and (D) number of frames per image of simulated visual snow, with corresponding flicker rate shown on the alternate y axis.

To assess reliability, we calculated intraclass correlation coefficients and 95% confidence intervals (CIs) for the composite measures that were averaged to produce estimates for an individual, which were 0.90 (95% CI = 0.47–0.97) for viewing distance, 0.78 (95% CI = 0.64–0.89) for separation, 0.92 (95% CI = 0.85–0.96) for luminance, and 0.80 (95% CI = 0.66–0.90) for frames per image, indicating good repeatability. Furthermore, repeat testing in an additional individual with VSS demonstrated stable estimates over a 5-week period, unaffected by the time of day (Fig. 5). 

Figure 5.

 

Repeat quantification of visual snow appearance in a single individual. The quantification task was repeated on 10 different days (every Monday and Thursday for 5 consecutive weeks) in the morning (grey symbols) and afternoon (black symbols) to obtain estimates for visual snow dot size (A), separation (B), luminance (C), and frames per image (D). Morning and afternoon sessions were approximately 5 hours apart, commencing within +/− 15 minutes of 9:30 AM and 2:30 PM. Parameter estimates were obtained using the procedure described in the methods and are plotted on the same scale as group data in Figure 4 (note +20 upwards shift in the y axis of panel D to capture range).

Participants generally reported good matches between the simulation and their own visual snow (Fig. 6A), except for a single person that was unable to complete the matching of flicker speed. Estimates and match ratings were unrelated for dot size, separation, and luminance (Kendall's tau correlations, P > 0.05) but a negative correlation between frames per image and match rating for flicker speed (τ_b = −0.41, 95% CI = −0.63 to −0.13, P = 0.01) indicated greater match satisfaction for those with faster perceived flicker. Satisfaction ratings for brightness matches in participants with visual snow that was purely achromatic (n = 15) or contained some color (n = 8) were similar (t(21) = 0.72, P = 0.48), as were ratings for single (n = 14) compared to multiple (n = 9) static types (t(21) = −0.31, P = 0.76). 

Figure 6.

 

Questionnaire results. (A) Match satisfaction ratings, spanning responses from simulated visual snow that matched their own visual snow not at all (0) to an exact match (10), for each appearance dimension considered in isolation. (B) Perceived severity ratings for each questionnaire item, with higher values indicating greater severity, for habitual visual snow in bright lighting (white filled boxes) and dim lighting (grey filled boxes). Box plots show the median (line), interquartile range (box), whiskers (extending to values within 1.5 times the interquartile range), and individual data (circles). (C) Percentage of participants reporting static in each color category in bright and dim lighting. (D) Percentage of participants reporting single or multiple static color types in bright and dim lighting.

The Relationship Between Visual Snow Appearance and Perceived Severity

Figure 6B presents the results of the perceived severity questionnaire. Self-rated visibility was greater for visual snow in dim compared to bright lighting (Z = 3.90, P < 0.001), as was the perceived interference with vision (Z = 3.74, P < 0.001), effort to ignore (Z = 3.63, P < 0.001), and degree of annoyance (Z = 3.59, P < 0.001), indicating that dim lighting worsens multiple aspects of daily visual experience. Visual snow color differed for bright and dim lighting in 48% of individuals, representing changes in color category (see Fig. 6C) or the number of color types (see Fig. 6D). 

Next, we investigated whether individual differences in the perceived visibility of visual snow were related to its measured appearance. The self-rated visibility of visual snow in dim lighting was linked to its estimated density, because individuals with higher visibility ratings had smaller separation estimates (τ_b = −0.41, 95% CI = −0.62 to −0.13, P = 0.01). However, perceived visibility was unrelated to estimated dot size (τ_b = 0.02, 95% CI = −0.27 to 0.31, P = 0.89), luminance (τ_b = −0.28, 95% CI = −0.53 to 0.01, P = 0.09), or frames per image (τ_b = −0.13, 95% CI = −0.41 to 0.17, P = 0.42). 

We also explored the influence of perceived visibility on daily visual experience. Regardless of lighting, visual snow rated as more visible interfered more with vision (bright: τ_b = 0.52, 95% CI = 0.28–0.71, P = 0.002; and dim: τ_b = 0.36, 95% CI = 0.08–0.59, P = 0.04) and took more effort to ignore (bright: τ_b = 0.35, 95% CI = 0.06–0.58, P = 0.04; and dim: τ_b = 0.48, 95% CI = 0.22–0.67, P = 0.005) but did not cause more annoyance (bright: τ_b = 0.19, 95% CI = −0.11 to 0.45, P = 0.26; and dim: τ_b = 0.17, 95% CI = −0.12 to 0.44, P = 0.31). 

Supplementary Table S1 details questionnaire responses for the 22 participants who felt that their vision was impaired in the dark. Visual snow was the most common contributor to this symptom (n = 19), followed by haloes (n = 16) and afterimages (n = 11). For the free text responses regarding difficulties with visual tasks in the dark, common themes were walking or navigating, distinguishing or identifying objects, driving, and reading or seeing fine details. 

Customized Test of Visual Interference: Confirming the Effect of Texture Scale in Controls

The quantification task revealed that visual snow is an extraordinarily fine static composed of tiny, closely packed dots. We simulated visual snow in controls to test our hypothesis that such a fine static likely interferes with the perception of equally fine textures but not textures of a much coarser spatial scale. 

Consistent with our hypothesis, log thresholds for fine and coarse textures were differentially affected by simulated visual snow. There was an effect of texture type (F(1, 22) = 7.08, P = 0.014, η_p² = 0.24) and an interaction between texture type and simulated visual snow (F(1, 22) = 4.98, p = 0.036, η_p² = 0.19; Fig. 7A) driven by elevated thresholds for fine compared to coarse textures with simulated visual snow (F(1, 22) = 9.49, P = 0.005, η_p² = 0.30). When viewed without simulated visual snow, thresholds for fine and coarse textures were similar (F(1, 22) = 1.27, P = 0.27). There was no overall effect of simulated visual snow on log thresholds regardless of texture type (F(1, 22) = 0.19, P = 0.67). Therefore, controls only required a stronger form signal (corresponding to an additional approximately 8% texture dots forming the concentric and radial patterns) for fine compared to coarse textures with simulated visual snow superimposed. 

Figure 7.

 

Performance on the customized test of visual interference. Log thresholds for discriminating global form for fine and coarse textures for (A) controls with and without external noise simulating visual snow and (B) VSS with textures individualized to their visual snow. Log spread of the psychometric function, which is inversely related to discrimination precision, for (C) controls with and without external noise simulating visual snow and (D) VSS with textures individualized to their visual snow. Mean (large symbols), 95% confidence intervals (error bars), and individual data (small symbols) are shown as diamonds in controls (left panels), colored dark blue for the conditions without simulated visual snow and light blue for conditions with simulated visual snow, and as red circles for participants with VSS (right panels).

Psychometric function slopes were unaffected by simulated visual snow, as there was no effect of simulated visual snow (F(1, 22) = 1.99, P = 0.17) on psychometric function spread (Fig. 7C) or interaction with texture type (F(1, 22) = 1.58, P = 0.22), suggesting that simulated visual snow did not make the task more difficult. However, slopes were shallower for fine textures, which had a wider spread than coarse textures regardless of simulated visual snow (F(1, 22) = 11.16, P = 0.003, η_p² = 0.34). 

We conclude that, if inherent visual snow acts like external noise, it will raise thresholds when individuals with VSS discriminate global form for textures matched in spatial scale to their own visual snow but not for coarser textures. 

Customized Test of Visual Interference: Individually Tailoring Textures to Visual Snow in VSS

We evaluated whether visual snow acts like external noise by comparing global form discrimination in VSS for fine and coarse textures that were individually tailored to each individual's visual snow. 

Log thresholds were comparable for individualized fine and coarse textures in participants with VSS (t(22) = 0.54, P = 0.60, d = 0.11, 95% CI = −0.30 to 0.52; Fig. 7B). Therefore, although inspection of Figures 7A and 7B indicates a slightly greater interindividual variation in thresholds in those with VSS compared to controls, texture spatial scale had a negligible effect on the strength of the form signal required in the VSS group as a whole. In contrast to the effects of external noise in controls, this result suggests that participants with VSS could successfully segregate their own visual snow from the global form signal contained in the texture stimuli, regardless of spatial scale. Unlike controls, slopes were unaffected by texture type in VSS as psychometric function spread was not wider for fine textures (t(22) = 1.44, P = 0.16, d = 0.3, 95% CI = −0.12 to 0.72; Fig. 7D). 

The main analysis assessed performance in control and VSS groups separately because physical stimulus differences between groups makes interpretation of a combined analysis challenging. However, for completeness, we include the results of an ANOVA comparing log thresholds between controls with simulated visual snow and participants with VSS. There was no effect of group (F(1, 44) = 0.37, P = 0.55) but an effect of texture type (F(1, 44) = 6.29, P = 0.016, η_p² = 0.13) due to elevated thresholds for fine textures. Although the interaction between texture type and group did not reach conventional statistical significance (F(1, 44) = 3.00, P = 0.09), the observed trend supported the main analysis as log thresholds were selectively elevated for fine textures in controls with simulated visual snow (F(1, 44) = 8.99, P = 0.004; M_difference = 0.13, 95% CI = 0.04–0.21) but not in participants with VSS (F(1, 44) = 0.30, P = 0.59; M_difference = 0.02, 95% CI = −0.06 to 0.11). 

Discussion

We quantitatively measured visual snow appearance using a custom designed matching task, revealing a consistent phenotype characterized by tiny, tightly packed dots. Our results advance understanding of visual snow, first by providing a more precise and interpretable account of its appearance than prior qualitative descriptions and second by identifying key features. In particular, visual snow was uniformly matched to external simulation dots in a narrow range of 5.2 to 6.9 arcseconds, indicating that dot size was a remarkably consistent feature of visual snow appearance across study participants. This clarifies the fine spatial scale of visual snow, which is difficult to truly appreciate from common comparisons to television static^1,21 and broad descriptors such as “small” and “tiny” dots in diagnostic criteria and rating scales.^1,5 Density was another salient feature, with more closely packed dots linked to greater perceived visibility. Dot luminance and flicker speed exhibited considerable heterogeneity, unrelated to individual differences in perceived visibility. They key features of dot size and separation can be measured in under 15 minutes. 

There has been recent interest in developing methods to quantify visual snow appearance using a matching task,²² similar to the longstanding approach in tinnitus research.²³ Whereas interpretation of qualitative descriptors in a rating scale is ambiguous, a matching task presents observers with the same external and modifiable reference. Our matching task was developed with input from an individual with VSS and confirmed as a successful approach in a small group of participants, who achieved satisfactory matches with good repeatability. Our approach differs from a recently published matching task,²² which used a different simulation (binary noise consisting of black and white pixels on a grey background) and task procedure (shorter maximum viewing distance, and simultaneous adjustment of dimensions). This alternate approach produced speed estimates that were similarly heterogeneous, but dot size was notably larger with greater individual variation in comparison to the consistently tiny dot size estimated in the present study. Methodological differences preclude direct comparison of density and brightness matches. Therefore, convergent evidence indicates that matching tasks are a viable means of measuring visual snow appearance, though further research may lead to optimization of the measurement approach. In particular, the unavoidable overlay of an individual's visual snow across the monitor may affect their percept of the simulation, altering the apparent size or density of simulation dots. An advantage of our approach was that separately matching each dimension of visual snow – rather than its overall appearance – to the simulation allowed observers to concentrate on a single feature. It also minimized potential interactions among apparent simulation dot size, brightness, and density, as participants could not adjust these dimensions simultaneously. Anecdotally, participants did not report any difficulty distinguishing between the simulation and their own visual snow that affected their ability to perform the task. 

The small size of visual snow dots presents another complication. First, reported estimates of visual snow dot size correspond to the external simulation and not the retinal image, which will be blurred by the eye's optics.²⁴ In this study, the average size match was a 6 second of arc simulation dot, which is expected to produce a retinal image a few minutes of arc in size, assuming an equivalent blur of 0.25 D (the standard tolerance for a clinical refraction²⁵). Second, visual snow simulations have also previously been used to estimate prevalence²⁶ and in neuroimaging research.^27–29 Images illustrating visual snow have been generated using text-to-image artificial intelligence (AI)³⁰ and a smartphone app.³¹ These previous depictions of visual snow have likely misrepresented dot size as participants in our study required long viewing distances (range = 8.2–11 m) for a single 0.28 mm pixel to match their visual snow dot size. Future simulations can easily achieve appropriately long viewing distances with a mirror to accurately represent visual snow, although this may not be feasible for typical magnetic resonance imaging (MRI) scanner setups or standard viewing of smartphone apps. Study participants should also have good visual acuity and optimal refractive correction where possible. 

Quantitative measurement approaches may also aid better assessment of both possible progression and the response to potential treatments. From previous research, we know that visual snow in people with VSS may date back to earliest memory or onset later in life,^1,13 but is typically unremitting⁴ and unresponsive to medication.^4,5 Furthermore, while the self-rated appearance of visual snow is stable in the short-term,⁵ people with VSS recollect either stable or worsening visual snow post onset.^1,13 Here, we demonstrated that estimated parameters remain consistent over a period of weeks in a single individual, indicating the method's potential for evaluating changes in visual snow appearance due to progression or in response to treatment. 

Furthermore, there is currently no validated questionnaire for assessing VSS severity. Although preliminary, our questionnaire revealed that visual snow is disruptive on many levels but its prominence only influences perceived detriments in function, not the level of annoyance. Visual snow was more prominent in dim lighting, with a corresponding increase in perceived interference with vision and effort to ignore, extending previous reports of difficulties in dim lighting.^1,5 Free text responses suggest that the increased visibility of visual snow at night interferes with navigating and distinguishing objects, whereas haloes around lights and afterimages likely affect nighttime driving. 

Visual snow is thought to correspond to spontaneous neural activity or noise in the visual pathway.^21,32,33 One possibility is that the visual cortex is hyperexcitable.^7,8,34 Electrical stimulation of the early visual cortex (V1, V2, and V3) produces still or moving phosphenes^35,36 and localized hyperfunction produces hallucinations that are phosphenes or visual snow-like percepts.³⁷ However, the level of internal visual noise is normal in VSS,^8,33 suggesting that spontaneous neural activity is not significantly elevated. Therefore, current research suggests that visual snow is unlikely to correspond to the perception of unusually high neural noise in the visual pathway. Alternatively, visual snow may arise from a failure to suppress typically subliminal levels of neural noise from awareness,³² with deficient cancellation of neural noise in the ascending visual pathway perhaps arising from disruptions in salience and default mode networks in people with VSS.²⁹ However, visual snow dot size constrains the likely origins of neural noise within the visual pathway that could give rise to the visual snow percept. Retinal signals with the highest spatial resolution are relayed to the visual cortex via the parvocellular pathway, beginning with foveal cones spaced approximately 0.5 arcmin apart³⁸ that transmit signals to midget ganglion cells with approximately 2 arcmin receptive fields.³⁹ Receptive field size increases with eccentricity^39,40 and along the visual pathway.⁴⁰ In the primary visual cortex, human population receptive fields in central vision are approximately 0.5 degrees,⁴⁰ with electrical stimulation producing phosphenes of comparable size.⁴¹ The retinal image of the single-pixel simulation dot is expected to be larger in size than a foveal cone but approaching the midget ganglion cell receptive field. Therefore, if we presume the visual snow percept simply corresponds to noisy activity at a particular level of the visual pathway, then it is unlikely to arise from the primary visual cortex but a retinal origin is possible. Elucidating the neural basis of visual snow, and any potential contribution of the early visual pathway, is a subject for further research. 

Furthermore, our results indicate that standard vision tests likely contain coarser spatial details than the visual snow percept. Compared to the visual snow simulation, the spatial detail of typical tests of signal extraction from noise^7,8 is an order of magnitude larger and the stroke of a letter on the 20/20 line of a visual acuity chart is 10 times wider. Therefore, we tested whether visual snow masks vision using dot textures matched in spatial scale. In controls, adding external noise simulating visual snow selectively impaired global form discrimination in textures of matched spatial scale, consistent with visual masking mechanisms.⁴² The observed degree of impairment was roughly comparable to the effect of healthy aging on global form discrimination in standard textures.¹⁰ Interestingly, in VSS, matching texture spatial scale to each individual's visual snow did not affect global form discrimination. Therefore, the effect of visual snow on texture perception is not simply equivalent to that of static external noise in the normal visual system. Future research is required to determine whether thresholds for other stimuli, such as dynamic textures, remain similarly unaffected. 

In conclusion, we developed a method for quantifying visual snow appearance, which revealed that tiny, closely packed dots are key features of the visual snow percept. We used this information to customize dot textures to the spatial scale of each individual's visual snow, revealing that visual snow does not obscure form perception like static external noise. Applications include monitoring progression, evaluating potential treatments, and guiding the development of targeted vision assessments. 

Acknowledgments

The authors thank Andrew Turpin (University of Melbourne, Australia) for assistance with software development and Philip Bedggood (University of Melbourne, Australia) for advice on retinal images. 

Disclosure: C.J. Brooks, None; Y.M. Chan, None; J. Fielding, None; O.B. White, None; D.R. Badcock, None; A.M. McKendrick, CentreVue iCare (F)