Red-green colour vision deficiency (CVD) affects ~ 4% of Caucasians. Notch filters exist to simulate CVD when worn by colour vision normal (CVN) observers (simulation tools), or to improve colour discrimination when worn by CVD observers (compensation tools). The current study assesses effects of simulation (Variantor) and compensation (EnChroma) filters on performance in a variety of tasks. Experiments were conducted on 20 CVN and 16 CVD participants under no-filter and filter conditions (5 CVN used Variantor; 15 CVN and 16 CVD used EnChroma). Participants were tested on Ishihara and Farnsworth-Munsell 100 hue tests, CVA-UMinho colour discrimination and colour naming tasks and a board-game colour-sorting task. Repeated-measures ANOVAs found Variantor filters to significantly worsen CVN performance, mimicking protanopia. Mixed-model and repeated-measures ANOVAs demonstrate that EnChroma filters do not significantly enhance performance in CVD observers. Key EnChroma results were replicated in 8 CVD children (Ishihara test) and a sub-sample of 6 CVD adults (CVA-UMinho colour discrimination and colour naming tasks) for a smaller stimulus size. Pattern similarity exists across hue for discrimination thresholds and naming errors. Variantor filters are effective at mimicking congenital colour vision defects in CVN observers for all tasks, however EnChroma filters do not significantly compensate for CVD in any.
Red-green colour vision deficiency (CVD) is a genetic condition that affects 8% of males and 0.5% of females in the Caucasian population[
Several compensation mechanisms might explain improved discriminability in CVD individuals, above that calculated theoretically from spectral sensitivity curves. More acute use of luminance cues often associated with hue differences, is possible both for red-green dichromats and anomalous trichromats. Dichromat compensation is especially salient when foveal and peripheral vision act together (for stimuli over 3°). Large-field trichromacy could be due to rod intrusion[
Despite availability of these many potential compensation mechanisms, the impact of CVD still reduces perception of a wide range of colours. The consequences of CVD impact on daily activities such as driving[
Aiming to improve lives of CVD observers, two recurrent questions in colour vision research arise. They are: (
The Variantor filter (Cambridge Research Systems Ltd., Rochester, UK) is the most known and widely available simulation filter. The logic underlying designs of simulation filters is to decrease colour contrast in normal trichromatic vision. To simulate protanopia and deuteranopia, the filter must block energy from parts of the spectrum that match the missing CVD cone type, so that the cone type simulated as missing in CVN observers, will have little or nothing to respond to. To simulate protanomaly and deuteranomaly, the filter must attenuate rather than block, certain parts of the spectrum. Degree and selectivity of attenuation would depend on severity of anomaly simulated. That is, if the spectral sensitivity curve of the abnormal cone is severely affected, attenuation will be high; if only slightly shifted, attenuation will be low. A previous study[
Compensation filters, or CVD aids, are notch filters that aim to reshape M and/or L cone spectra to increase differences between peak spectral energies, so increasing colour contrast and providing more distinguishable responses arising from photoreceptors. This feature is particularly useful for anomalous trichromats, whose main characteristic is excessive overlap of spectral sensitivity profiles between M and M′ cones or L and L′ cones (usually 12 nm or smaller for anomalous trichromats, instead of about 15 nm between M and L cones in normal trichromacy)[
The current study aims to measure and describe the effects of simulation (Variantor) filters in CVN observers and compensation (EnChroma) filters on colour perception in CVD and CVN observers. In addition to two clinical colour vision tests (Ishihara and FM100) also used in previous studies[
Research data for this study were collected in the United Kingdom, at Anglia Ruskin University (ARU) and in Portugal, at the University of Minho (UMinho). Twenty colour normal (CVN) and 16 red-green colour defective (CVD) adult participants performed experiments under two conditions: no-filter and filter (Variantor or EnChroma). The colour normal participants were 13 females and 7 males with an average age (± SD) of 23.35 ± 4.60 years. Five CVN participants (2 females and 3 males with an average age of 20.80 ± 3.70 years) performed tasks with Variantor filters. Fifteen CVN participants (11 females and 4 males with an average age of 24.20 ± 4.66 years) performed tasks with and without EnChroma filters. Normal colour vision was confirmed with the Ishihara test[
Ethics approval was obtained from the Faculty of Science and Engineering Research Ethics Panel (VHS DREP 0917-01 for adults; FSTFREP 15538 for children) at ARU, and the University of Minho Ethics Committee for Research in Life and Health Sciences (SECVS 175/2017, CEICVS 004-2019 and CEICVS 052-2021) in line with the ethical principles of Helsinki declaration of 1975. All participants, as well as parents/guardians in the case of children, were provided with verbal and written explanations about the experiments. Written informed consent (and assent in the case of the child) was obtained before experiments were conducted.
Two types of notch filters were used: the simulation filter, Variantor (Cambridge Research Systems Ltd., Rochester, UK) and the compensation filter, EnChroma CX-65[
Graph: Figure 1 Transmittance spectra for filters and spectral power distribution for lighting, screen guns and sorting pieces. Average transmittance spectra (in arbitrary units) of the filters (red lines for Variantor; blue lines for EnChroma; solid lines for filter to right eye, dotted lines for filter to left eye) in panels (a), (b) and (c) and average spectral power distribution (SPD): of the D65 light source (black dashed line) in panel (a); of screen guns of the Sony-GDM F520 (red dashed line for red gun; green dotted-dashed line for green gun and dark blue solid line for blue gun) in panel (b); and of the sorting pieces (black solid line with white circles for white pieces; blue solid line with blue circles for blue pieces; orange solid line with orange triangles for orange pieces; red solid line with red circles for red pieces; brown solid line with brown squares for brown pieces; green solid line with green crosses for green stimuli) as radiance in panel (c). Error bars indicate ± 1 standard error of the mean (SEM) across three measurements of filter transmittance and of the spectral power distribution (SPD) of the D65 light. If not visible, error bars in terms of repeatability of measurements from 400 to 730 nm are within the line/symbol size.
Whenever the simulation or compensation filters were needed to perform a colour discrimination task, they were worn over the top of the observer's habitual refraction, which was checked to be optimal by an optometrist. As per EnChroma instructions, participants had a 10-min adaptation period prior to testing for both Variantor and EnChroma filters. Participants performed the tasks with and without filter (Variantor/EnChroma) in counterbalanced order across task and filter conditions. Colour discrimination tasks were carried out on the British adult participants with and without compensation/simulation filters twice, on different days, to gain an estimate of repeatability variance magnitude.
The effect of filters was assessed for five colour vision tasks. Two were conventional clinical colour discrimination tests (Ishihara and FM100), two were computerised colour vision discrimination and naming tasks, and the final task was a sorting task with coloured board-game pieces. All tests are described below.
The clinical colour vision tests and the sorting task, which required an external light source, were performed in a JUST Normlicht LED box (JUST Normlicht GmbH; Weilheim, Germany) set to a metameric D65 at 100% brightness with uniform grey walls and an approximately uniform spatial distribution of the light source at Anglia Ruskin University. In Portugal a custom-built light box, with Munsell N5 paint (VeriVide Limited, Leicester, United Kingdom) and a D65 simulator light source (VeriVide BS 950 PT 1, F20 T12/D65, VeriVide Limited, Leicester, United Kingdom), was used. The dotted black line in Fig. 1a represents the average of three measurements of the spectral power distribution (SPD) of the D65 light source used in the Anglia Ruskin University light box.
Standard versions of the Ishihara Test for Colour Deficiency (Kanehara Trading Inc, Tokyo, Japan, 38 plates, 2011[
A standard edition of the Farnsworth Munsell 100 Hue Test (FM100)[
The colour vision assessment task by the University of Minho (CVA-UMinho, Colour Science Lab, Centre of Physics, University of Minho, Portugal) is a computer-based colour discrimination task, in which a coloured square target must be located against a grey background. The neutral grey background during stimulus presentation consisted of static luminance noise discs of variable diameter. The luminance of each disc was randomly selected from a predefined luminance range, to preclude colour discrimination judgements being based on luminance cues. A coloured square target appeared either to the left or the right side of the screen, the chromaticity of which was added to the grey background discs. The participant's task was to identify the position of the square target, which had a different hue to the background. Target hue saturation commenced at 100% and followed a staircase paradigm until the chromaticity appeared to match that of the background for the participant. Below threshold level, the participant would be able to identify the position of the target only by chance. This task is based on similar principles to that of the Cambridge Colour Test (CCT) and The Colour Assessment and Diagnosis (CAD) test, however all parameters could be modified according to researcher needs. This task was also modified for the colour naming experiment. Further details of the CVA-UMinho colour vision discrimination task can be found elsewhere[
In these experiments, both at Anglia Ruskin University and the University of Minho, the CVA-UMinho stimuli were presented on a CRT colour computer screen (Sony-GDM F520 at Anglia Ruskin University and Sony-GDM FW900, Sony Corporation, Tokyo, Japan; see Fig. 1b for the gun's spectra of the Sony-GDM F520), driven by a computer graphics system (Visage MKII, Cambridge Research Systems Ltd., Rochester, UK) and calibrated in luminance and colour using a telespectrophotometer (SpectraColorimeter, Model PR-650, Photo Research Inc., Chatsworth, California, USA). The participant's viewing distance to the computer screen was 1 m, resulting in a field of view of 17°, with the stimulus chromatic square (5° in size) presented on a neutral grey background of luminance discs (0.17° to 0.51° size). The room was otherwise completely dark. When the task was conducted, the participant identified to which side of the screen (L or R) the square target appeared using a simple joy-stick control. Before testing started, participants adapted for two-minutes to the darkness of the room with a mean luminance grey background on the screen (mean luminance = 11 cd/m
Sixteen hues were tested, 6 corresponding to hues on dichromatic (protan, deutan and tritan) confusion lines. The other 10 hues were spaced approximately equally around the neutral point (the grey uniform background), which was CIE illuminant D65 (chromaticity coordinates: u′ = 0.1947 and v′ = 0.4639) expressed on the CIE 1976 UCS chromaticity diagram. For each hue, a colour discrimination threshold was estimated by computing the Euclidean distance between the selected point and the neutral point. Thresholds were determined after 4 staircase reversals, or after presenting the stimulus for 25 trials, whichever occurred first. The last 15 trials contributed to threshold estimation where a minimum of 2 reversals had occurred.
The CVA-UMinho task was modified to present 5° chromatic squares on the neutral grey luminance noise background (and 1° squares for the participant subsample), for the same 16 hues (10 evenly spaced hues, plus the 6 hues corresponding to dichromatic confusion lines), but for this experiment squares were drawn close to the centre of the screen to reduce positional uncertainty of the stimulus for naming. Saturation was fixed at 100%, 66% and 33%. Each hue-saturation combination was presented 3 times, which resulted in 144 trials (16 hues × 3 saturations × 3 presentations). Participants were instructed to name aloud each square presented, with one of the 11 basic colour terms (BCT: monolexemic abstract colour names whose extensions are not included in other basic terms, which are used consensually and consistently in a language)[
Four sets of coloured, wooden board game pieces (red, orange, white and blue) from the Settlers of Catan 5th Edition[
For almost all statistical analyses carried out in this project, participant data were grouped either as from CVN or CVD groups. This is because the literature produced about Variantor and EnChroma filters suggests that filters simulate, or compensate for, congenital red-green colour vision deficiency (CVD). This grouping does not affect overall outcomes (see "Discussion" section on "Specific effects of EnChroma filters on anomalous trichromats versus dichromats"). For all repeated-measures analyses of variance (ANOVAs) conducted in this study, when more than 2 levels of factor were present (such as 16 levels of hue in the CVA-UMinho discrimination and naming tasks), family-wise adjustments to degrees of freedom were made according to the strict Greenhouse–Geisser correction when testing for statistical significance.
British adult participants performed each task twice to gain information about the repeatability of our measures. Two types of analyses were performed. First for the Variantor filters in the CVN group, repeated measures ANOVAs were conducted for repeat (first run, second run) and filter (no-filter, Variantor) within-subject factors for different task measures. Second, for the EnChroma filters, three-way mixed-model ANOVAs were conducted with repeat (first run, second run) and filter (no-filter, EnChroma) as within-subject factors, and participant group (CVN, CVD) as the between-group factor (except for the sorting task, as CVN participants did not perform the task with EnChroma filters). For the sorting task, a separate three-way repeated measures ANOVA was conducted on the CVD group with within-subject factors of repeat (first run, second run), sorting task (by colour, by shape) and filter (no-filter, EnChroma).
Analyses of both Variantor and EnChroma data did not reveal any significant effect of repeat on Ishihara (error scores), FM100 (TES), CVA-UMinho discrimination thresholds, CVA-UMinho naming scores or sorting task results, nor any significant repeat interactions with other factors (all Ps > 0.175). There were significant main effects of filter and participant group, as well as significant interactions with other factors in these analyses, however as these effects were also found in the main analyses to follow (using the average of the two runs), these are reported below. Means and standard errors across the two runs are provided in Table S1 in the supplementary information.
A one-way repeated measures ANOVA with filter (no-filter, Variantor) as the within-subjects factor was conducted to test for the effect of Variantor filters on Ishihara error scores (red bars in Fig. 2a). A significant effect of filter (F
Graph: Figure 2 Mean Ishihara and FM100 TES (total error scores) scores, with and without filters. Average error scores (excluding misreadings) on the Ishihara test (a) and average TES scores on the FM100 test (b) for CVN-Variantor (red), CVN-EnChroma (grey), CVD-EnChroma (blue) adult conditions and CVD Children EnChroma (yellow) condition without filter (solid fill with skinny borders) and with filter (translucent fill with thick borders). Horizontal dashed lines represent cut-off points to "fail" the Ishihara (a, 3 errors) and the FM100 (b, 100 TES). Error bars indicate ± 1 standard error of the mean (SEM). *(P < 0.05), **(P < 0.01), ***(P < 0.001) and n.s. (not significant) indicate significance levels from the ANOVA tests or Tukey post-hoc comparisons.
Six one-way repeated measures ANOVAs with filter (no-filter, Variantor) as the within-subjects factor were conducted to test for the effect of Variantor filters on six FM100 indicators (TES, red bars in Fig. 2b; SQR, Angle, C-Index, S-Index and time) as very different numerical scales are used for each indicator. These analyses revealed significant effects of Variantor on TES, SQR and C-Index, a near-significant result for Angle, but no significant effects on S-Index, or time score (see Table 1 for details). Using the FM100 TES cut-off point of 100, as is used clinically, Variantor filters worn by every CVN participant would have resulted in them being diagnosed as CVD.
Table 1 FM100 Indicator scores, with and without filters. Significant main effects and Tukey post-hoc comparisons on filter vs no-filter effects for FM100 indicators.
Variantor CVN (n = 5) F test EnChroma CVN (n = 10)a EnChroma CVD (n = 16)a No filter Variantor Significance level No filter EnChroma No filter EnChroma TES 68.00 ± 40.22 210.00 ± 30.62 31.00 ± 12.24 44.13 ± 11.19 140.85 ± 9.97 125.77 ± 9.11 SQR 7.34 ± 2.55 14.40 ± 1.08 5.34 ± 0.66 6.58 ± 0.64 11.66 ± 0.53 11.04 ± 0.50 C-index 1.49 ± 0.29 2.67 ± 0.15 1.20 ± 0.10 1.36 ± 0.09 2.10 ± 0.08 2.00 ± 0.08 Angle 57.41 ± 3.80 29.32 ± 3.75 58.21 ± 3.44 55.32 ± 6.80 23.16 ± 3.64 36.44 ± 4.73 S-index 1.42 ± 0.06 1.74 ± 0.13 1.28 ± 0.03 1.39 ± 0.03 1.50 ± 0.04 1.50 ± 0.03 Time (s) 443.33 ± 97.68 491.83 ± 23.65 452.80 ± 38.20 441.95 ± 34.57 365.50 ± 36.91 370.46 ± 38.80
Exact P-values and n.s. (not significant) indicate statistical significance levels of main ANOVA.
A two-way repeated measures ANOVA with filter (no-filter, Variantor) and hue (16 hues tested) was conducted to test the effect of Variantor filters on discrimination thresholds measured using the CVA-UMinho task with 5° stimuli (Fig. 3a). This analysis revealed significant main effects of filter (F
Graph: Figure 3 Mean discrimination thresholds for CVA-UMinho task, with and without filters. Discrimination threshold measures for the CVN-Variantor (a), the CVN-EnChroma (b) and the CVD-EnChroma (c) conditions without (solid) and with (dotted) filters with 5° (big circles along black lines) and 1° (small triangles along grey lines) stimuli. Marker colours are approximations of hues presented at maximum saturation. Vertical lines represent the two extremes of the protan (red), deutan (green) or tritan (blue) confusion lines. Error bars show ± 1 standard error of the mean (SEM). ***Indicates filter vs no-filter thresholds are significantly different by Tukey post-hoc testing (P < 0.001).
To show Variantor filter effects on proportions of colour naming categories used (or basic colour terms: BCT), we calculated hit and error scores (in percent) per BCT using the modal criterion. Confusion matrices (11 × 11 BCTs) were created for CVN participants with and without Variantor filters and these are pictorially represented in Fig. 4a–f. For ease of comparison, in this pictorial, the proportion of colour names given without filter is shown by the inner pie; the proportion given with filter, are shown in the surrounding doughnut. Numerical CVN-Variantor confusion matrices behind these "pies" and "doughnuts" are provided in Supplementary Table S3 for the modal criterion (see Supplementary Table S4 for numbers generated using the loose criterion). Due to stimulus characteristics (isoluminant stimuli at 11 cd/m
Graph: Figure 4 Pie pictorials for naming results of British participants using CVA-UMinho task, with and without filters. Naming task percentages for the modal criterion for CVN-Variantor (a–f), CVN-EnChroma (g–l) and CVD-EnChroma of the British sample with 5° stimuli (m–r) and CVD-EnChroma with 1° stimuli (s–w). Inner pies (solid colours) show naming proportions without filters and outer doughnuts (translucent colours) show them with filters. Notes: In colour key, NF stands for non-filter condition and F(V/E) stands for filter (Variantor/EnChroma) conditions. Percentages represented can be found in Supplementary Tables S3, S5, S9 and S11. Equivalent graphs for the Portuguese sample can be found in Supplementary Fig. S1. For example, hit-rate (i.e. 88% without and 45% with Variantor for Green in panel (a); 87% without and 5% with Variantor for Blue in panel (b); error-rate (i.e. 5% without and 20% with Variantor for Yellow, as Green in panel (a) appear in Supplementary Table S3. Pie chart colours are approximations of best exemplars of each basic colour term (BCT).
A one-way repeated measures ANOVA with filter (no-filter, Variantor) as the within-subjects factor was conducted on naming hit scores. This analysis revealed a significant main effect of Variantor filters (F
Specific colour naming error scores by CVN using Variantor filters are illustrated across hue in Fig. 5a using a format that allows comparison with hue discrimination data of Fig. 3a: higher values denote worse performance in both. A similar pattern of results is seen across hue when comparing colour discriminability thresholds (Fig. 3) and colour naming results (Fig. 5). It was not possible to conduct ANOVA analyses on naming scores across hue for CVN participants wearing Variantor filters as results for some hues lacked variance. For example, CVN mean hit scores for 126° no-filter were 100.00 ± 0.00%; mean hit scores for 4.66° and 184.66° (both protan axes) 342° and 347.26° (a deutan axis) with filter were 0.00 ± 0.00%. The results of Fig. 5a demonstrate that CVN participants wearing Variantor filters have higher error scores for naming all hues, except for 306°. This might be due to the peak transmittance for Variantor filters being in the region of greenish hues (see peak from 540 to 600 nm; red line in Fig. 1), leading to perceptual enhancements for complementary hues and a higher use of the modal BCT, purple.
Graph: Figure 5 Mean naming results using CVA-UMinho task, with and without filters. Average naming error scores (100%—hit score%) on the CVA-UMinho colour naming task for the CVN-Variantor (a), the CVN-EnChroma (b) and the CVD-EnChroma (c) conditions without (solid) and with (dotted) filters with 5° (big circles along black lines) and 1° (small triangles along grey lines) stimuli. Marker colours are approximations of hues presented at maximum saturation. Vertical lines represent the two extremes of the protan (red), deutan (green) or tritan (blue) confusion lines. Error bars show ± 1 standard error of the mean (SEM).
Mean sorting error scores with and without Variantor filters are shown in Fig. 6 (red bar of Fig. 6a; zero error scores without filter). Differences in error scores between colour sorting and shape sorting for the real-world task for each participant were compared with a repeated measures ANOVA. More errors were found with Variantor filters (5.40 ± 3.31; mean without filter was 0.10 ± 0.10), however statistical significance was not reached (P = 0.189), as three of the five participants had zero errors with and without Variantor filters. When proportional changes in time taken for sorting into groups of colour versus groups of shape for each participant (see red bars in Fig. 6b) were compared, sorting by colour took approximately twice as long as sorting by shape (expected as there were six colour groups and three shape groups). When Variantor filters were worn, significantly longer colour sorting times than shape sorting times were found than when they were not worn (F
Graph: Figure 6 Mean sorting task error and timing results, with and without filters. Average difference between colour and shape error scores (a) and average proportion between colour and shape time scores (b) on the sorting task for the CVN-Variantor (red bars) and the CVD-EnChroma (blue bars) not using (solid with skinny border) and using (translucent with thick border) the filter in the sorting task. Error bars show standard error of the mean (SEM). *(P < 0.05) and n.s. (no significant) indicate overall effects from the ANOVAs.
Adults using EnChroma filters were tested in two different research laboratories (Anglia Vision Research, UK; Colour Science Lab, Portugal). To check that participants in both laboratories produced equivalent results on our tasks, a mixed-model ANOVA was conducted with laboratory (British, Portuguese) and filter (no-filter, EnChroma) as within-subject factors, and participant group (CVN, CVD) as the between-group factor. For the sorting task, a separate analysis was conducted on the CVD group only, as CVN adults did not perform this task with EnChroma filters. These analyses did not reveal any significant effect of laboratory, or any interaction between laboratory and measures for Ishihara (error scores), FM100 (TES), CVA-UMinho discrimination thresholds and CVA-UMinho naming (5°) hit scores, or sorting task (error scores): all Ps > 0.134. Any significant effects of filter and participant group are described in the main analyses to follow, in which data from both laboratories are combined (more details on specific laboratory analyses are provided in the Supplementary information).
A two-way mixed-model ANOVA with filter (no-filter, EnChroma) as the within-subjects factor and adult participant group (CVN, CVD) as the between-subjects factor was conducted to test for the effect of EnChroma filters on Ishihara test error scores (see Fig. 2a, grey bars for CVN, blue bars for CVD). This analysis revealed a significant overall effect of filter (F
For the CVD child sample, a one-way repeated measures ANOVA, with filter (no-filter, EnChroma) as the within-subjects factor was conducted on Ishihara error scores (yellow bars in Fig. 2a). This analysis did not reveal a significant effect of filter (F
Six two-way mixed-model ANOVAs with filter (no-filter, EnChroma) as the within-subjects factor and participant group (CVN, CVD) as the between-group factor, were conducted to test for the effects of the EnChroma filters on each of the six FM100 indicators listed in Table 1 (see Fig. 2b for representation of TES results, grey bars for CVN, blue bars for CVD). As expected, the overall effect of participant group (CVN vs CVD) was significantly different for all but the time score indicator (P = 0.635), i.e., TES [F
For all four indicators, EnChroma filters led to a trend of worse scores (increased for TES, SQR and C-Index; decreased for Angle) compared to when no filter was worn in the CVN group, but for the CVD group, EnChroma filters led to improved scores (decreased for TES, SQR and C-Index; increased for Angle), compared to when no filter was worn (see this cross-over trend for TES in Fig. 2b).
The wearing of EnChroma filters did not influence clinical diagnoses with FM100 for any participant. According to the test scoring criteria in which a TES < 16 suggests superior discrimination, 16–100 is average discrimination and > 100 is low discrimination[
A three-way mixed-measures ANOVA with filter (no-filter, EnChroma) and hue (16 hues tested) as within-subjects factors and participant group (CVN, CVD) as the between-group factor, was conducted to test effects of EnChroma filters on discrimination thresholds measured on the CVA-UMinho task with 5° stimuli (see Fig. 3b for CVN, circle markers in 3c for CVD). This analysis revealed significant main effects of hue (F
A subsample of 6 CVD participants also performed the CVA-UMinho discrimination task with a smaller stimulus size (1° versus 5° for main experiment). A three-way repeated-measures ANOVA with size (5°, 1°), filter (no-filter, EnChroma) and hue (16 hues tested) as within-subjects factors was conducted to test the effects of EnChroma filters (see Fig. 3c, big circles for 5°, small triangles for 1°). This analysis revealed a significant interaction between size and hue (F
Results for CVN versus CVD groups have demonstrated a lack of effect of EnChroma filters on hue discrimination. On average for the CVD group, the differences in discrimination thresholds without and with filters (NF-F) are very small, much smaller than the effect that Variantor filters had on discrimination thresholds for the CVN group. The magnitudes of effect of EnChroma filters in the CVD group are on average, too small to be of functional significance. Although participant data were usually grouped into CVN and CVD groups as previously mentioned, one question of interest could be, are EnChroma filters helpful for any specific CVD group? We plot differences in discrimination thresholds with and without filters in Fig. 7 for specific CVD groups. The grey area in Fig. 7 reveals the variance (± 1 SD) across the CVD group, which is higher than the measured differences in discrimination thresholds due to the filter itself in most cases and higher than repeatability of the measures themselves in CVD participants (average ± 1 SD of 0.0005).
Graph: Figure 7 Mean no filter minus filter discrimination thresholds for CVA-UMinho test. Differences between no filter and filter discrimination threshold were calculated for the four CVD types using EnChroma (red solid-line for protanopia, red dashed-line for protanomaly, green solid-line for deuteranopia and green dashed-line for deuteranomaly) and for the CVN groups using Variantor (black dotted-line) and EnChroma (black dashed-dotted-line). Values above the solid horizontal black line across hues at 0 (y-axis) indicate improvements in discrimination with the filters whilst values below, indicate degradations in discrimination. The grey area indicates variance (± 1 SD) across the CVD group. Marker colours are approximations of hues presented at maximum saturation. Vertical lines represent the two extremes of the protan (red), deutan (green) or tritan (blue) confusion lines. Error bars show ± 1 standard error of the group mean (SEM).
A three-way ANOVA with filter (no-filter, EnChroma) and hue (16 hues tested) as within-subjects factors, and participant group (CVN, Protanope, Protanomalous, Deuteranope and Deuteranomalous) as the between-group factor on discrimination thresholds was conducted (thus excluding Variantor data). This analysis showed significant main effects of CVD group on discrimination thresholds (F
Do EnChroma filters significantly change the appearance of colours in CVD observers, such that presented hues might be given a different colour name? To estimate EnChroma filter effects on proportions of colour naming categories (or basic colour terms: BCT) used for different hue angles, we calculated hit and error scores using the modal criterion. Confusion matrices (11 × 11 BCTs) were created for CVN and CVD participants with and without EnChroma filters and these are pictorially represented in Fig. 4g–r for British participants (and in Supplementary Fig. S1a–p for Portuguese participants). These data were not collapsed into a single figure because the British sample provided six BCTs as modal responses for presented stimuli (green, blue, pink, purple, orange, brown), whereas the Portuguese participants provided eight (the same as the British, plus red and grey). CVD-EnChroma confusion matrices are provided in Supplementary Table S9 for British participants (Supplementary Table S10 for loose criterion) and in Supplementary Table S11 for Portuguese participants (Supplementary Table S12 for loose criterion).
On examining Fig. 4m–r (and Supplementary Fig. S1i–p and Supplementary Tables S9 and S11) for CVD participants, EnChroma filters only slightly changed modal hit scores for all BCTs for British and Portuguese CVD participants. Using a criterion of ≥ 3% change (much smaller than the 20% criterion used for Variantor filter effects): British CVD participants (3 Protan, 6 Deutan), made more errors in naming blue (− 6%) and improved (fewer errors) in naming brown (+ 6%) and pink (+ 5%); Portuguese CVD participants (5 Protan, 2 Deutan) made more errors in naming blue (− 4%) and pink (− 10%) and fewer errors in naming purple (+ 3%), grey (+ 6%), orange (+ 20%) and brown (+ 3%). Changes also occurred in naming for CVN participants wearing EnChroma filters (Fig. 4g–l and Supplementary Tables S5 and S9). With EnChroma filters, British CVN participants made more naming errors for orange (− 5%) and purple (− 3%) and fewer errors for naming brown (+ 4%); Portuguese CVN participants made more naming errors for red (− 13%), grey (− 7%) and brown (− 3%) and fewer errors for naming blue (+ 3%) and orange (+ 7%). An additional statistical analysis on BCTs is provided in the Supplementary information.
As the same hue angles were tested in both discrimination and naming tasks, specific colour naming error scores for each hue angle (modal criterion), with and without EnChroma filters are provided for CVN and CVD groups in Fig. 5b,c. These data were analysed using a three-way mixed-measures ANOVA with filter (no-filter, EnChroma) and hue (16 hues tested) as within-subjects factors, and participant group (CVN, CVD) as the between-group factor. There was a significant effect of hue on naming hit scores (F
CVN participants showed improvements (reductions in naming errors by ≥ + 3%) with EnChroma for 98.5° (tritan axis) and 198°. CVN participants increased naming errors (by ≥ − 3%) at 167.3° and 347.3° (deutan axes) and 306°. CVD participants showed improvements with EnChroma (≥ + 3%) for 18°, 54°, 98.5 (tritan axis) and 306°, but increased naming errors (≥ − 3%) for 126°, 162°, 184.66° (protan axis), 198°, and 347.5° (deutan axis). Additional analyses on BCTs and on differences in hit-scores (i.e., no-filter minus filter) are provided in the Supplementary information.
For the subsample of 6 British CVD participants who performed the CVA-UMinho naming task with a small stimulus size (1°), EnChroma changed modal hit scores for all 6 BCTs (see Fig. 4s–w and Supplementary Table S13; for loose hit scores see Supplementary Table S14). For this size, using a criterion of ≥ 3% in hit scores: CVD participants improved in naming green (+ 3%), but worsened for blue (− 4%), purple (− 4%) and brown (− 4%). Supplementary Tables S13 and S14 provides details of increases and reductions in naming errors for CVD with 1° stimulus size.
Specific CVD colour naming error scores across hue for 1° stimulus size, with and without EnChroma filters are provided in Fig. 5c to allow comparison with colour discrimination data of Fig. 3c. An ANOVA with filter (no-filter, EnChroma), size (5°, 1°) and hue (16 hues tested) as within-subjects factors, was conducted to test for effects of EnChroma filters and size on naming hit scores (using the modal criterion). This analysis revealed significant main effects of size (F
CVN observers using EnChroma filters did not perform this task. For CVD participants, sorting error scores (Fig. 6a, blue bars) and proportional increases in sorting times when sorting pieces by colour versus shape (Fig. 6b, blue bars) were analysed with a one-way repeated measures ANOVA with filter (no-filter, EnChroma) as the within factor (see Fig. 6a,b, blue bars). No significant effect of filter on sorting error scores was found (P = 0.349), but there was a significant increase in colour relative to shape sorting times for CVD participants with EnChroma (F
Variantor and EnChroma notch filters aim to simulate and compensate for CVD, respectively (see Fig. 1). Whilst Variantor filters are successful in mimicking protan colour vision deficiency in CVN observers, our results show that EnChroma filters do not significantly improve colour vision discriminability in clinical tests or laboratory tasks, do not significantly affect colour appearance such that different colour naming categories are used, and do not improve performance accuracy or speed on a real-world colour sorting task, although findings might be different with long-term filter usage[
Previous research[
A previous study[
If we look qualitatively at discrimination thresholds with the CVA-UMinho task, neglecting statistical significance, any threshold reduction seen in one part of the spectrum (such as at 162°) is compensated for by a threshold increase in another part of the spectrum, such as at (184.66° or 198°). Concurrent results appear in the colour naming task: improvements for certain BCTs (e.g., brown and pink for CVD; brown for CVN; 184.66°) were accompanied by more errors in others (e.g., blue and orange for CVD; orange and purple for CVN; 306°). Patterns of quantitative improvement and degradation between discrimination thresholds and colour naming are also not clear, but the effects of wearing EnChroma filters are very small, even for small sized (1°) stimuli where their effects could be potentially greater, given the almost exclusive zone of red-green processing by the fovea and the poorer discrimination overall thresholds providing greater opportunity for improvement if it exists. Such small changes in discrimination with EnChroma filters are not likely to contribute to differences in colour perception (although a possibility remains that findings could be different if tested after very long term exposure[
Compensation filters aim to increase colour contrast by increasing differences in peak spectral sensitivities of the viewed image, used by existing cone types in the observer. This strategy may work for anomalous trichromats, where overlap of spectral absorptions of two cone pigment types is greater and peaks closer, compared to normal trichromatic vision[
In dichromatic vision, notch filters (like EnChroma filters) cannot increase colour contrast as only 1 cone type exists in the range of the notch. However, the notch can result in changes in luminance contrast for dichromats. In addition, the ideal position of the notch is likely to vary, even within the same CVD classification (e.g. protanomalous). Apart from the variability in the sensitivity overlap of the L and M cones existent in anomalous trichromacy[
Our results show no statistically significant differences on FM100 results, measures of hue discrimination and naming in laboratory tasks, or on performance on a real-world colour sorting task in CVD observers with versus without EnChroma filters, although sorting-by-colour took significantly longer in time for CVD participants when wearing them. Our sample size of CVD observers is modest (n = 16 CVD: Protanopia n = 4; Protanomaly n = 4; Deuteranopia n = 4; Deuteranomaly n = 4), although our results were also replicated in 8 children (Ishihara test) and for smaller sized stimuli (CVA-UMinho discrimination and naming tasks). Would we have found statistically significant differences if we increased our sample size? Fig. 7, which shows discrimination thresholds studied intensively in the laboratory, helps to answer this question. EnChroma filters showed significant losses of hue discrimination for the group of protanopes at a protanopic confusion line (Hue angle 4.66°; P = 0.002) and a significant enhancement for deuteranopes near a deuteranopic confusion line (Hue angle 162°, P = 0.026). We do not believe that the effects found for dichromats are due to increases in colour contrast but are more likely due to changes in luminance contrast. For the protanope, the luminance difference usually experienced between medium and long wavelengths was potentially reduced by the filter, leading to losses in hue discriminability. For deuteranopes, the filter potentially introduced new luminance differences not present without the filters, leading to improved hue discriminability. No significant differences were found for discrimination thresholds with and without the EnChroma filters for anomalous trichromats. The effects here were smaller than the variance (± 1 SD) of measures within the CVD group at specific hues, too small and variable to be of functional value. If a study were designed to seek statistical significance in anomalous trichromats for the greatest but very small enhancement effects (e.g. at 0.0052 ± 0.0052 at 162° for deuteranomalous and at 0.0061 ± 0.0040 184.66° for protanomalous participants), sample size estimates[
As noted above, EnChroma filter effects on colour discrimination thresholds (CVA-UMinho task) are greater for dichromats, than for anomalous trichromats, suggesting that altered luminance cues (rather than chromatic cues) were used in judgements. The luminance noise in the CVA-UMinho task is greater (Michelson contrast of 24 ± 4%) than in the Ishihara test (18 ± 1%), and changes randomly from trial to trial, increasing its power. Whereas a significant effect of EnChroma filter on error scores was found on the Ishihara test in adults with CVD (see Fig. 2a), it was not found for CVA-UMinho discrimination thresholds (Fig. 3c). This suggests that when potential new luminance cues are disguised by higher noise levels, virtually no effect of EnChroma filters on hue discrimination thresholds remains.
Any red-green CVD compensation filter can introduce different colour cues and specially luminance cues, which might result in improved clinical test scores (such as in the Ishihara test and the FM100) in CVD observers, and improved discriminability for some hues (but degraded discriminability for others). They might even help CVD observers in specific colour-combinations important in everyday tasks. However, the use of filters must be restricted to specific tasks, as our results also demonstrate that they may hinder other judgements and may lengthen times for completing some real-world tasks. To date, no study has demonstrated enhanced overall colour discrimination for persons with red-green colour vision deficiency when using compensation filters.
Some of this work was presented as a poster at the European Conference on Visual Perception (ECVP) in 2020. This work was supported by an Evelyn Trust Grant (to SJW), HEFCE QR (Quality Related) Funds (to Anglia Vision Research) to support a Postdoctoral Research Fellow (LA) and visits (LA and JMML) to the laboratories of Anglia Vision Research at ARU, to facilitate completion of this project. This work was also supported by the Portuguese Foundation for Science and Technology (FCT) in the framework of the Strategic Funding UIDB/04650/2020. The authors thank Andreia E. Gomes (PhD student supported by the Portuguese Foundation for Science and Technology (FCT), Portugal), Ashley Gray (a research assistant supported by the QR fund), Emily Mailman and Laura Douds (who were undergraduate Optometry students) for their help with data collection.
All authors (i.e. L.A., J.M.M.L., M.A.F., S.J.W.) contributed to the study design. Data collection was performed by L.A and J.M.M.L. Data analysis was performed by L.A.L.A. wrote the first version of the manuscript. All authors participated in redrafting and critically revising the manuscript. All authors approved the final version of the manuscript for submission.
The data that support the findings of this study are available from the corresponding author upon reasonable request.
The authors declare no competing interests.
Graph: Supplementary Information.
The online version contains supplementary material available at https://doi.org/10.1038/s41598-022-13877-9.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
By Leticia Álvaro; João M. M. Linhares; Monika A. Formankiewicz and Sarah J. Waugh
Reported by Author; Author; Author; Author