What does visual snow look like? Quantification by matching a simulation

Abstract

The primary symptom of visual snow syndrome (VSS) is the unremitting perception of small, flickering dots covering the visual field. VSS is a serious but poorly understood condition that can interfere with daily tasks. Several studies have provided qualitative data about the appearance of visual snow, but methods to quantify the symptom are lacking. Here, we developed a task in which participants with VSS adjusted parameters of simulated visual snow on a computer monitor until the simulation matched their internal visual snow. On each trial, participants (n = 31 with VSS) modified the size, density, update speed, and contrast of the simulation. Participants’ settings were highly reliable across trials (intraclass correlation coefficients > 0.89), and they reported that the task was effective at stimulating their visual snow. On average, visual snow was very small (less than 2 arcmin in diameter), updated quickly (mean temporal frequency = 18.2 Hz), had low density (mean snow elements vs. background = 2.87%), and had low contrast (average root mean square contrast = 2.56%). Our task provided a quantitative assessment of visual snow percepts, which may help individuals with VSS communicate their experience to others, facilitate assessment of treatment efficacy, and further our understanding of the trajectory of symptoms, as well as the neural origins of VSS.

Introduction

People with visual snow continuously perceive a veil of small flickering dots covering the visual field. The symptom is surprisingly common, affecting an estimated 2.7% to 5.5% of the population (Kondziella, Olsen, & Dreier, 2020). Visual snow is the primary symptom of the less common visual snow syndrome (VSS), which has a 1% to 3% prevalence (Kondziella et al., 2020). VSS includes additional symptoms such as photophobia, poor night vision, and floaters. VSS is significant because it is persistent (Graber et al., 2022; Mehta, Garza, & Robertson, 2021; van Dongen, Waaijer, Onderwater, Ferrari, & Terwindt, 2019; Yoo, Yang, Choi, Kim, & Hwang, 2020), can be debilitating (Eren, Ruscheweyh, Straube, & Schankin, 2020; Solly, Clough, Foletta, White, & Fielding, 2021), and lacks effective treatments (Eren & Schankin, 2020; Lukáčová, Mastík, & Minks, 2018; Puledda et al., 2022). 

Visual snow symptoms have been characterized by qualitative surveys, but more quantitative measures of symptoms are lacking. Questionnaires have revealed that visual snow is typically black and white, small, “moderately” dense, “moderately” fast, and dynamic (Puledda, Schankin, & Goadsby, 2020; Puledda et al., 2021; Viana, Puledda, & Goadsby, 2020). However, these qualitative descriptions lack standardized visual references, making them difficult to compare across individuals who may interpret the labels differently. Additionally, these questionnaires have only included a few possible rating levels, limiting their precision. 

Establishing a method to quantify the visual snow symptom will advance the field in many ways. First, such a method may be used to measure the efficacy of treatments; Traber, Piccirelli et al. (2020) noted that the current lack of such a method may hinder clinical trials. A method of quantifying visual snow could also help with diagnosis by helping differentiate visual snow from other conditions (Hang, Leishangthem, & Yan, 2021). Additionally, quantifying visual snow is valuable for several research purposes. Investigators could use such a method to determine whether there are stable individual differences and whether patients with visual snow may be subgrouped by the appearance of their visual snow. Quantification methods could also facilitate assessment of the time course of the symptom after onset and aid identification of its neural origins. 

Other internally generated perceptual experiences have been quantified by matching. For example, research in the auditory domain shows that participants with tinnitus can consistently and efficiently estimate the sound quality, frequency, and loudness of their percept by adjusting the settings of a simulation (e.g., Neff, Langguth, Schecklmann, Hannemann, & Schlee, 2019). Matching techniques have also been useful in estimating the spatial frequency, orientation, and contrast of illusory gratings seen after observing high contrast grating patterns (e.g., Georgeson, 1980). 

Here, we tested a similar method to measure the appearance of visual snow, including element size, density, update speed, and contrast. To accomplish this, we generated a simulation of visual snow that participants could alter until it matched their individual visual snow symptom. Our main goals were to determine whether this method was capable of simulating visual snow reliably, to investigate how consistent settings were across individuals, and to provide measurements of visual snow. 

We show that our visual snow simulation task provides an effective method for quantifying visual snow symptoms. Our matching results reveal that visual snow is, in general, very small, relatively fast, low contrast, and broadband in spatial frequency. 

Methods

Participants

Three groups of participants were recruited: five participants with VSS (VSS small group), five participants without visual snow (control group), and 29 participants with VSS (VSS large group). The purpose of the small groups was to determine the feasibility of our methods. Each participant in the control group was randomly matched with a different participant from the VSS small group. The results from the individual VSS participants in the small group were used to generate references for the controls to match with the same technique. The small VSS and control groups each consisted of two participants with psychophysics experience and three naïve observers. Studying the VSS large group allowed us to examine whether our findings from the VSS small group replicated in a larger sample with stricter criteria for excluding comorbidities. Three participants were in both the VSS small group and VSS large group (total of 31 participants with VSS). 

Inclusion criteria for the small group were less strict regarding comorbidities because we did not expect comorbidities to interfere with our assessment of the feasibility of the task. More strict inclusion criteria were applied to the VSS large group to reflect standards in the existing VSS literature. Eligibility for the VSS small group was determined by the following inclusion criteria: 18 to 80 years old, ability to comply with study instructions, and self-report of visual snow covering the visual field for at least the past 3 months. Although additional visual symptoms were not required for inclusion, all participants in the VSS small group met the criteria for the full syndrome (at least two of the following visual symptoms: afterimages, blue field entoptic phenomena, trails or trailing behind moving objects, light sensitivity, or poor night vision). Participants were ineligible for the VSS small group for any of the following: lack of fluency in English, a diagnosed or self-reported intellectual disability, severe central nervous system disease, any physical problem that would impede data collection or interpretation (e.g., visual field loss), a current acute episode of psychosis (assessed with self-report and the Brief Psychiatric Rating Scale; Overall & Gorham, 1962), or an ophthalmological condition that could cause visual snow symptoms. Control participants had to meet the same eligibility criteria as the VSS small group, with the addition that they were excluded if they had ever experienced visual snow. 

Participants in the VSS large group had similar inclusion criteria to the VSS small group with the following exceptions, which were intended to more closely align with prior VSS literature: 18 to 60 years old, corrected Snellen acuity of 20/25 or better (assessed at 100 cm), met criteria for visual snow syndrome (endorsed continual, pan-field visual snow for at least 3 months, and at least two of the following visual symptoms: afterimages, blue field entoptic phenomena, trails or trailing behind moving objects, light sensitivity, or poor night vision). Several additional exclusion criteria for the VSS large group included substance dependence within the last 12 months, head injury with skull fracture or loss of consciousness for more than 30 minutes, visual anomalies (e.g., strabismus, cataracts), or hallucinogenic substance use (LSD, psilocybin, peyote, DMT, ayahuasca, PCP, ketamine, dextromethorphan, or salvia divinorum) within the last 12 months or 12 months prior to the onset of VSS symptoms. To minimize possible confounds with migraine status, we recruited comparable numbers of participants with and without migraine in the visual snow syndrome groups (Table 1). Participants were recruited through a review of medical records, word of mouth, University of Minnesota websites, and local visual snow social media group posts. 

Table 1.

 

Demographic data for the controls, VSS small group, and VSS large group are shown. Snellen visual acuity values are reported as decimal fraction (e.g., 20/20 = 1).

 

All procedures conformed to the tenets of the Declarations of Helsinki and were reviewed and approved by the University of Minnesota Internal Review Board. Written informed consent was obtained prior to study participation. Participants were compensated for their time. 

Displays and stimuli

Simulated visual snow was presented using MATLAB (MathWorks, Natick, MA) and Psychtoolbox-3 (Kleiner, Brainard, & Pelli, 2007) on an iMac desktop computer (Apple, Cupertino, CA) connected to a FlexScan SX2462W monitor (60-Hz refresh rate, 1920 × 1200 resolution, 53 × 33-cm screen size; Eizo, Ishikawa, Japan) in a dark room. The simulated visual snow was generated by assigning a percentage of pixels to a random draw from a binary luminance distribution (snow pixels) and assigning the rest to mean gray (background pixels). A Gaussian blur (SD = 1 pixel) was then applied to more closely mimic the appearance of visual snow. A subset of pixels updated on every frame, with each pixel luminance assignment lasting for the duration determined by the speed parameter. Pixel updates were distributed evenly throughout this duration, so pixels appeared to change continuously over time (as opposed to all pixels updating simultaneously after a certain number of frames). 

Simulated snow was presented on a uniform mean gray field (80.8 cd/m²) within a 10° wide square region that was centered 6° to either the right or left of center (counterbalanced across participants) and outlined with a 2-pixel black border. An identical square outline was presented on the other side for comparison purposes. For participants with VSS, this comparison region was blank (i.e., mean gray), whereas for controls it contained a second simulated snow comparison stimulus (Figure 1). The background surrounding the squares was mean gray. 

Figure 1.

 

Visual snow matching task. (a) Participants sequentially set the size, density, update speed, and contrast of the visual snow and could then adjust the density, update speed, and contrast simultaneously to make their final settings. They repeated these five steps 10 times. (b) Visualization of how changing the density and contrast affected the appearance of the simulation. All demonstration images are zoomed in 10×.

Procedure

Participants compared the appearance of the simulated snow in the target square on one side of the screen to the appearance of the reference square on the other side. Participants were free to move their eyes back and forth between the two squares as needed. For participants with visual snow, the reference square was mean gray, matching the background. This allowed them to judge the appearance of the simulated snow compared to their visual snow symptom. Observers sequentially adjusted the size, density, update speed, and contrast of the visual snow simulation to match the appearance of the snow in the reference square. 

For control participants, the reference square showed a second visual snow simulation (i.e., comparison stimulus) generated using the mean settings obtained in this task from an individual participant in the VSS small group; each control participant was randomly paired with a different individual in the VSS small group and asked to match their mean settings. 

All participants practiced the task at least once before beginning experimental trials. Participants completed 10 trials, with each trial consisting of the five steps described below. Each step was preceded by instructions. During the step, participants made adjustments while the simulated snow displays were presented continuously until they were satisfied with their setting. 

Participants first adjusted the size of the visual snow elements. Because participants reported their visual snow was smaller than an individual snow element at the smallest size that could be displayed on our monitor at a typical viewing distance, participants adjusted the size of the visual snow by changing their viewing distance from the monitor. Distances ranged between 70 cm (front of the table with the display monitor) and 227 cm (back of the room). A staff member measured and recorded their distance from the screen. The size of the target and reference squares was adjusted to be 10° visual angle across viewing distances; this was limited to a viewing distance of 130 cm due to the size of the monitor, such that the comparison squares were smaller when viewed from farther than 130 cm (n = 22, so the mean visual angle across all 29 participants was 7.8°). During the size adjustment step, the visual snow simulation was displayed at the median density, maximum contrast, and a dot lifetime of 0.033 seconds to ensure that the simulation was easily visible. Control participants were positioned at the same distance (the mean distance of the VSS participant to which they were assigned) on every trial. 

Next, participants adjusted the density of the simulated snow. This and the remaining three parameters were set with button presses on a keyboard. The density settings controlled the proportion of pixels that were assigned to be randomly drawn visual snow elements versus background. Density could be set to 20 different log-spaced values ranging from 0.5% to 100%. During density settings, the starting density was set to a random value; the contrast was always 100%, and the update speed was always 30 Hz. 

Participants then adjusted the update speed of the simulation, with density remaining as previously set. The update speed setting determined the duration of an individual snow element on the screen. As noted above, the updating of the elements was offset so an equal number of elements were replaced on each frame. We used 30 update speed steps ranging from 2 Hz (30 frames, 0.5-second element lifetime) to 60 Hz (1 frame, 0.0167-second element lifetime). 

Participants then adjusted the contrast of the snow elements. As noted above, snow elements were assigned luminance values from a binary distribution centered at mean gray. The contrast parameter controlled the standard deviation of this distribution (i.e., the luminance of the light and dark elements) prior to blurring. Participants could set the contrast to 20 log-spaced values ranging from 0.39% to 100%. Contrast was adjusted last so that low visibility settings would not interfere with the ability to match the other parameters. 

After making initial adjustments to each parameter separately, participants could then make any final changes to the density, update speed, and contrast simultaneously using different keypresses for each parameter. Mid-way through the experiment, participants were asked how well the simulation matched their visual snow and whether they would change anything about the simulation to make it more accurate. 

Analysis

Data were analyzed in MATLAB. Distance measurements were used to convert the size of the snow elements to degrees of visual angle. Because informal observations suggested that similar-looking simulations could be achieved by changing either the density or contrast settings, we also calculated the root mean square (RMS) contrast for the noise patterns as a measure of total contrast energy in the simulations using the following equation:  

\begin{eqnarray*} RMS = \sqrt {\frac{{\mathop \sum \nolimits_{i = 1}^n {{\left( {\frac{{{\rm{\;}}{L_i} - {\rm{\;}}\bar L}}{{\bar L}}} \right)}^2}}}{n}} \end{eqnarray*}

where n is the number of pixels in the image, L_i is the local pixel luminance, and \(\bar L\) is the mean luminance of the image (Kukkonen, Rovamo, Tiippana, & Näsänen, 1993). 

We quantified the consistency of all parameter settings across trials using a two-way mixed, average measures intraclass correlation coefficient (ICC) for consistency calculated as (MS_R – MS_E)/MS_R, where MS_R is mean square rows and MS_E is mean square error (Koo & Li, 2016). We calculated the percent error in control settings (relative to the mean settings of the assigned VSS participant) to determine their task accuracy and consistency. 

To characterize the appearance of snow across the population, we calculated the mean of settings across the large VSS group for each parameter. To estimate the variability in appearance across individuals, we calculated coefficients of variance for each setting. Using the mean settings, we also calculated the average spatial frequency spectrum for our noise simulations by generating 1000 sample images, computing the spectrum of each using a Fourier transform, and taking the mean across samples. To investigate potential differences in settings between individuals with VSS with a history of migraine from those with no history of migraine, we used an omnibus repeated-measures analysis of variance (ANOVA) with factors for participant (random variable), migraine status (fixed), and visual snow measure (e.g., size and speed, as a fixed variable). Data were log transformed prior to the ANOVA to satisfy the assumption of normally distributed data. Finally, we performed a thematic analysis of participants’ subjective reports on the task to identify potential ways to improve the simulation in future studies. Thematic analysis codes were assigned by two team members independently and then compared, and differences were reconciled. 

Results

Participants with VSS adjusted the size, density, update speed, and contrast of a visual snow simulation to match their visual snow symptom. Ten matches (trials) were set in a single session. 

Observers can reliably match the appearance of their visual snow

The displays simulated observers’ visual snow effectively. First, we found that participants made consistent settings across repeated trials (Figure 2), which we quantified using ICCs for consistency (Table 2). In our VSS small group, all ICC values were in the excellent reliability value range (i.e., above 0.9). In fact, the settings by the VSS small group were as consistent or more so than those made by control participants matching physically presented noise (Table 2). To determine whether the effectiveness of the task generalized to a larger group of participants with VSS, we tested 29 participants with VSS who met stricter inclusion criteria than the VSS small group. Again, participants’ settings were highly consistent across trials (Table 2). This suggests that the task effectively simulated visual snow; it is unlikely that observers with VSS would have responded so consistently if the adjustment variables could not replicate the appearance of their particular symptom. Additionally, when we asked participants how our procedures could be improved (see further discussion below and Supplementary Table S2) many participants specifically said that their matches were good representations of their visual snow—some even wished they could show the simulation to others. 

Figure 2.

 

Snow matching settings from the VSS small group (upper) and control group (lower). Observers with VSS adjusted their viewing distance to match the size of the simulation to their visual snow. All participants adjusted the update speed, density, and contrast of the simulation. The RMS contrast of the visual snow images was calculated post hoc to account for the perceptual relationship between density and contrast. Individual settings are shown, from left to right, for size (a), update speed (b), density (c), contrast (d), and RMS contrast (e). Open colored circles represent settings on an individual trial and the means are shown as gray closed circles for each participant. Error bars show standard error of the mean. For control participants, the true value of the simulated reference snow they were attempting to match is plotted with a black bar. Note that this bar is covered by the data points for Control 2 on the contrast plot (d, lower).

Table 2.

 

ICCs for size, update speed, density, contrast, and RMS contrast settings in each group.

 

Next, we determined how accurately the control participants could match the adjustable (target) snow to the reference snow, which used the mean values from the assigned participant in the VSS small group. The control participants also adjusted the variables with high consistency across trials, except for density, where the consistency was poor (Figure 2, Table 2). We also examined possible biases in control participants’ settings by calculating the mean percent difference between the adjusted and reference parameters. The errors in control participants’ settings were not consistently made in any one direction for any of these parameters. The average error in matching the update speed was 17%, but the percent error for contrast and density were higher (31% and 88%, respectively; Supplementary Table S1). We hypothesized that the low consistency in density settings and poor accuracy for density and contrast settings could be due to a perceptual tradeoff between contrast and density. To address this possibility, we examined the mean RMS contrast of the simulated snow in each trial, which reflects a combination of density, contrast, and blur. Control participants’ average percent error in setting the RMS contrast was much lower and their settings were extremely consistent (10% error; ICC = 0.989). 

Visual snow contains high temporal frequencies and a broad range of spatial frequencies, and is moderately low contrast

We used average setting values (across participants) to characterize the typical appearance of visual snow as captured by our task (see Discussion for possible differences in other contexts). 

Size

The individual visual snow elements were very small, with an average size (pre-blurring) of 0.0103° (0.620 arcmin or 37.2 arcsec; Figure 3, Table 3). Note that the actual size of our visual snow elements was increased somewhat by blurring with a Gaussian with 1 pixel SD. Nevertheless, many participants positioned themselves at the back of the room, making the elements as small as possible (i.e., minimum of 24.9 arcsec), and several commented that they would have moved back further if they were able. There was moderate variability in the size settings across individuals (coefficient of variation = 35.1%), though this may have been affected by the room size limitation. 

Figure 3.

 

Mean settings of participants in the VSS large group. Each dot represents the mean setting of an individual with VSS. Boxplots depict the median setting (middle bar), 25% and 75% quartiles are the upper and lower edges of the box, and 1.5× the interquartile range is plotted with a dashed line (whiskers). From left to right, settings are shown for size (a), update speed (b), density (c), contrast (d), and RMS contrast (e).

Table 3.

 

Visual snow simulation settings.

 

Update speed

The mean update speed (snow update frequency) was relatively rapid at 18.2 Hz (Table 3). There was high consistency in update speed settings within individuals, and variability across participants was relatively large (coefficient of variation = 80.2%). 

Density

On average, density settings were low, with a mean of 2.87% of pixels being snow elements, prior to blurring (Table 3). Settings appeared relatively variable across participants (coefficient of variation = 100.4%), due in part to the low average density (i.e., most were between 1% and 5%). As expected, blurring increased the actual number of pixels that visibly differed from the background luminance (see Figure 4). 

 

Figure 4.

 

Example video of simulated visual snow with mean density and contrast settings. To accurately illustrate the element size, this video should be viewed with a frame rate of at least 60 Hz and at least 833 × 833 pixels at full size (17.2 cm width) from a distance of 166 cm. Video is available on the journal website.

Contrast

Settings for contrast were relatively high (mean contrast = 63%) (Figure 3, Table 3). This was moderately consistent across individuals (coefficient of variation = 19.5%). 

RMS Contrast

RMS contrast was generally low, with an average of 2.56% (Table 3). This value is roughly 10% of “full contrast” simulated snow: The RMS contrast of images simulated at full density and 100% pixel contrast was around 29%, due to blurring. There was moderate variability in RMS contrast settings across individuals (coefficient of variation = 22.5%) (Figure 3). It should be noted that there is a large difference between the contrast and RMS contrast settings because the contrast value refers to the standard deviation of only the pixels assigned as snow elements (vs. mean gray), prior to blur. RMS contrasts are lower, because they are calculated after blurring and include the pixels assigned to background mean gray (when the density of snow elements is less than 100%). 

Together these parameters combined to produce simulated snow that was low contrast and broadband in spatial frequency. An example generated with the mean settings from the VSS large group is shown in Figure 4. The spatial frequency spectrum of simulations with the average settings was roughly Gaussian (due to the Gaussian blur) and fell to half its maximum at 19.8 cycles per degree (Supplementary Figure S1). We found no evidence for significant differences in the settings according to migraine status in the VSS large group (repeated-measures ANOVA): main effect of migraine status, F(1, 108) = 2 × 10⁻⁵, p = 1.0; interaction between visual snow measure and migraine status, F(4, 1305) = 0.64, p = 0.6 (Supplementary Figure S2). 

Qualitative analysis suggests simulated snow matches internal percepts

Mid-way through the trials, participants were asked to comment on whether they would change anything to make the simulation more accurately match their visual snow. Several common themes emerged from their responses. The most common was that the simulation matched well (16 responses of 29 participants queried). The next most common theme was that their visual snow appeared faint on our mean gray screen compared to other viewing conditions (14 responses); many participants specifically noted that their visual snow was much more visible on the dark wall behind the screen. Several participants also mentioned wishing they could control the color of the simulated visual snow (nine responses). Another common comment (six responses) was that the simulation should capture other VSS symptoms (e.g., afterimages, floaters, colorful waves). Five participants said the pattern of movement of the dots did not match their visual snow. Other suggestions had fewer than four responses and are listed in the Supplementary Results (Supplementary Table S2). 

Discussion

We quantified the appearance of visual snow using a novel simulation matching task. This allowed us to estimate important aspects of the appearance of visual snow, including the size, update speed, density, and contrast of visual snow elements for both individuals in, and the average of, a group of VSS participants. We found that the task was effective—participants reported that the simulation was generally good at capturing the appearance of their snow, and their responses were consistent across trials. Participants with VSS were on par with or even more consistent than controls, who used the task to match physically presented snow. Our observation that settings were made consistently supports the notion that visual snow is not a confabulation or due to malingering, as it would be difficult to achieve such consistency if participants were not able to reference a stable internal percept. 

Additionally, our matching task provides the best estimate to date of what visual snow looks like. We found that visual snow is generally composed of elements that are small and quick, and that the combined percept is broad in spatial frequency and of moderate to low contrast. The matching task may be clinically valuable in diagnosing visual snow and for monitoring changes in severity over time. 

Relation to prior work

Our findings agree with and extend previous literature that attempted to characterize visual snow. In perhaps the most thorough of prior studies, participants filled out a 30-day symptom diary in which they rated some of the same qualities we measured with our task (i.e., density, update speed, size). Their median size response was “small,” the median density response was “Moderate static. Some masking of vision,” and the median speed response was “moderate” (Puledda et al., 2022). Many other studies have described snow percept as “small” or “pixelated” (Bou Ghannam & Pelak, 2017; Sampatakakis et al., 2022; Schankin, Maniyar, Digre, & Goadsby, 2014; Traber, Piccirelli et al., 2020; van Dongen, Alderliefste, Onderwater, Ferrari, & Terwindt, 2021), and indeed the proposed criteria for visual snow syndrome describes the symptom as “dynamic, continuous, tiny dots in the entire visual field” (Puledda et al., 2020). Our results agree with these qualitative descriptions and build on them by quantifying the average element size (with blur) to be less than 2 arcmin and an update speed of around 18 Hz (±5.5). 

Two prior attempts have been made to simulate visual snow with the goal of helping users communicate their experience to doctors and others. Perri, Simonetti, Gervasi, and Amato (2022) made a mobile app that simulated visual snow with a few options for setting the color, intensity (i.e., density), size, and luminance contribution. Similar visual snow simulations were developed at VisionSimulations.com. Neither has been used to quantify snow symptoms in a population for research purposes. Our simulation also provides finer sampling of the range of possible settings, which should allow for more precise communication of one's visual snow symptom. Note added in proof: we just became aware of a newly-published report also using a simulated snow matching method to characterize visual snow in detail (Brooks et al., 2024). 

Matching simulations with symptoms has also proven valuable in the auditory domain, where it has been successfully used to personalize sound therapy for tinnitus. Historically, tinnitus matching has included frequency and loudness dimensions (for a review, see Hazell, 2009) and also more recently timbre (Henry, 2016). Participants with chronic tinnitus made consistent settings with low variability when using a method of adjustment, analogous to our methods (Neff et al., 2019). Recent sound therapy treatments were shown to be more effective when personalized with matching (Wang, Tang, Wu, Zhou, & Sun, 2020). Having a precise method to simulate visual snow may similarly be useful in individualizing therapies to relieve visual snow, including our own recent use of adaptation to visual noise (Montoya, Mulder, Lee, Schallmo, & Engel, 2023). 

The neural origins of visual snow remain unclear and under investigation (e.g., Metzler & Robertson, 2018; Puledda, Schankin, Digre, & Goadsby, 2018; Traber, Aldusary et al., 2020). Based on our conclusion that visual snow is low contrast and contains high temporal frequencies, it is tempting to speculate that it arises along the magnocellular visual pathways. However, there is substantial overlap between magnocellular and parvocellular responses in the temporal frequency range (mean 18 Hz) our participants reported (Skottun, 2016). Furthermore, a subset of participants commented that their visual snow appeared colorful, in agreement with previous surveys (Puledda et al., 2020; Schankin et al., 2014), which points to the involvement of the parvocellular and/or koniocellular streams. The small element size, generating relatively high spatial frequency content, also suggests the involvement of the parvocellular pathway. Determining the pattern of cortical activity that could produce a percept similar to what we have measured, given what is already known about the response properties of visual neurons in different visual areas, is a promising direction for future research. 

Contextualizing visual snow measurements

Although prior reports have indicated that visual snow elements are small (e.g., Puledda et al., 2022; Sampatakakis et al., 2022; Schankin et al., 2014; van Dongen et al., 2021), our quantification reveals just how small they are. We found that they may approach psychometrically measured limits of spatial resolution, which is approximately 1 arcmin for an average human eye with 20/20 acuity for discriminating high-contrast stripes from a uniform mean field (Yanoff & Duker, 2009). With the blur, the diameter of an isolated snow element that is notably brighter or darker than the background is about 1.86 arcmin in diameter prior to any additional blurring produced by the optics of the eye. For reasons described below, we believe the simulation was visible to participants despite its proximity to acuity threshold. Due to the limits of our simulation (e.g., room size, blurring), it is also possible that visual snow elements could be even smaller than we measured. 

We found that visual snow is relatively fast, with a mean temporal frequency around 18 Hz. Most neurons in V1 and V2 of non-human primates respond maximally to temporal frequencies below 8 Hz (Foster, Gaska, Nagler, & Pollen, 1985). Intriguingly, our visual snow update speed is very similar to the reported flicker frequency of migraine aura (17.8 Hz, SD = 9.7) as measured with an adjustable flickering LED (Crotogino, Feindel, & Wilkinson, 2001). It is possible that similar mechanisms may underlie both percepts, as has been suggested previously given the high comorbidity of migraine in VSS (Klein & Schankin, 2021; Schankin et al., 2014; White, Fielding, Pelak, & Schankin, 2022). However, it should be noted that VSS is distinct from migraine and that migraine medications generally do not improve VSS symptoms (Mehta et al., 2021; Puledda et al., 2022; van Dongen et al., 2019). 

The mean RMS contrast setting of 2.56% indicates that perceived contrast of visual snow was relatively low. In healthy controls, detection thresholds of static white noise stimuli on a uniform background are around 3% of the background (standard deviation of normally distributed pixel contrasts) (Fairchild & Johnson, 2007; Xiao, Farrell, & Wandell, 2005). RMS is also a calculation of the standard deviation of pixel contrasts, so by that measure the contrast of the simulated snow was close to the reported threshold. 

A number of additional factors likely increase the visibility of our simulated snow, however. Perhaps most importantly, the snow is dynamic, with some pixels updating on every frame. Although threshold measurements have not been made, to our knowledge, for dynamic white noise, our simulation (Figure 4) is clearly more visible when animated. In addition, the luminance distributions in our snow simulations were not normal, primarily due to the density parameter that constrains a portion of the image to be equal to the mean luminance. This increased the local contrast at dense patches and/or isolated elements within the display increasing their visibility. That the snow simulations were above threshold visibility is further supported by the fact that the control participants could match the replayed visual snow settings. 

Limitations and future directions

Perceptual matching studies to estimate parameters of intrinsic experiences are limited by the fact that the intrinsic experience is still present during the matching process. In order to achieve an ideal match, one might expect participants to adjust the simulation to be blank, resulting in the perception of only internal visual snow in both target and reference squares. In pilot testing, however, most participants said that either they could segregate their snow from the simulation (e.g., by color) or their snow was suppressed by the simulation. Visual snow is generally thought to be less visible on patterned backgrounds (Metzler & Robertson, 2018). Thus, it is possible that, when the simulation was sufficiently strong, it suppressed the visibility of the superimposed snow. Alternatively, some amount of contrast from the visual snow may have been internally combined with contrast of the simulation. If so, this might cause our simulated snow contrast values to underestimate the true perceived contrast of visual snow. The size, speed, and density parameters would likely be less affected, as they would be visible regardless. 

For simplicity, we chose a limited number of intuitive parameters for participants to control. When prompted for suggestions, several additional variables were mentioned to improve the accuracy of the simulation, including adding a dimension for color (n = 9) and control over the amount of blur (n = 2). Future work could also give participants controls in the Fourier domain (e.g., temporal and spatial frequency center and bandwidth). We also limited the range of values that participants could set, and some participants wished they could have gone beyond these boundaries (e.g., wanted to move farther than the back of the room or set the contrast higher than the maximum). 

People with VSS report perceptual uniformity of the snow across the visual field, but this uniformity remains empirically untested. Our participants freely viewed the target and reference squares, shifting their eyes between the two, and so were likely matching foveal or parafoveal percepts. Future work could adapt our paradigm for peripheral measurements. 

It is likely that the mean gray background reduced the visibility of visual snow. Fourteen participants specifically commented that it was relatively difficult to see their visual snow on the screen or that it was easier to see in other situations such as the dark wall behind the monitor. These anecdotes are supported by previous reports that visual snow is worse in both very bright and dark environments (e.g., Bou Ghannam & Pelak, 2017; Lauschke, Plant, & Fraser, 2016; Puledda et al., 2022; Solly et al., 2021). Future studies could repeat the present method with a range of luminance backgrounds. 

Our data make it clear that visual snow differs substantially among individuals. The high ICC values, along with the relatively wide range of settings across individuals, indicate reliable differences among the participants. For example, some see snow elements that update around 4 Hz, but for others visual snow updated at 60 Hz. Because we found that density and contrast settings were interdependent and errors in the controls’ settings for density and contrast were fairly high, the potential individual differences we observed in density and contrast settings among people with VSS should be interpreted with caution. RMS settings appear likely to be a more valid measure than density or contrast alone using our simulation method. The full population variability may differ from our findings, for at least two reasons. First, the visual snow symptom specifically tends to become less noticeable or bothersome over time (Graber et al., 2022), and most of our participants had experienced visual snow for a long time. The mean duration of visual snow in our VSS large group was 14.4 years at the time of the study, with six participants having visual snow for as long as they could remember (mean age at data collection = 28 years for these six participants). Whether visual snow becomes less visible over time (perhaps lower density/contrast) or whether people simply pay less attention to it over time remains to be tested. Second, and conversely, our recruitment methods may have biased our sample toward individuals who have sought medical care for visual snow and may therefore have captured a subset of individuals with more intrusive visual snow. 

Visual snow simulation matching is a promising avenue for future research and clinical applications. Studies using the matching method could test whether visual snow is more visible in certain environments or viewing situations. This in turn could help people with VSS manage their symptoms more effectively. Longitudinal studies could use matching as an assessment for the effectiveness of treatments, which is currently lacking but necessary for clinical trials (Traber, Piccirelli et al., 2020). The visual snow simulation could also aid in diagnosis, as it may help distinguish a participant’s symptoms from other conditions (e.g., Hang, Leishangthem, & Yan, 2021; Lauschke et al., 2016; Puledda et al., 2018; Yoo et al., 2020). 

Acknowledgments

Supported by a University of Minnesota Office of the Vice President for Research Grant-in-Aid (574483); grants from the National Eye Institute, National Institutes of Health (F31 EY034016 and T32 EY025187, a vision training grant); a grant from the Diversifying the Community of Neuroscience Program (R25 NS117356); a grant from the National Institutes of Health (UL1TR0024940); and a National Science Foundation Research Training Grant in translational and sensory science (DGE 1734815 at the University of Minnesota). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or The University of Minnesota. 

Commercial relationships: none. 

Corresponding author: Samantha A. Montoya. 

Email: [e-mail]. 

Address: Graduate Program in Neuroscience, University of Minnesota, Minneapolis, MN 55455, USA.