Evaluation of native canine skin color by smartphone‐based dermatoscopy

Abstract

Background

Human skin color, predominantly determined by the chromophores of melanin, hemoglobin, and exogenous carotenoids, is often measured to serve various medical and cosmetic applications. Although colorimetry has been used to evaluate the skin erythema in allergic dogs, the native canine skin color remains unknown.

Methods

We measured the skin color in 101 healthy dogs using a calibrated optical system with a smartphone and a mobile dermatoscope DermLite DL1. The results were retrieved in the CIELAB color system and compared to the human color ranges.

Results

The lightness (L*) of canine skin ranged from 28.5 to 78.3, which is slightly broader than that of human skin. There was a difference of 3.9 in redness (a*) between canine and human skin, but this variation could be attributed to the similarly valued colorimetric error of the optical system. Nonetheless, the skin yellowness was significantly different for dogs and humans (respective median b* of 12.3 versus 16.6, p < 0.01). This difference might be due to canids not being able to accumulate typically yellowish carotenoids. Furthermore, the native canine skin color did not exhibit a typical dependence between the coordinates of lightness (L*) and yellowness (b*), known as the individual typology angle, °ITA.

Conclusion

We reported the first dataset of the native canine skin color in the CIELAB color space. We discovered a similarity in skin lightness and a difference in skin yellowness. However, further studies are needed for a more precise comparison of skin redness.

1. INTRODUCTION

Researchers in the cosmetic industry and medicine have shown a great interest in evaluating the color of human skin. Knowing the complete skin color range serves diagnostic purposes, skin‐care products, facial recognition applications, graphic arts, and prosthetics. ¹ , ² Human skin color is largely controlled by individual genetic heritage, which was shaped geographically according to the distribution of ultraviolet radiation. ³ Three main chromophores determine the human skin color: melanin, hemoglobin, and exogenous carotenoids. Melanin, normally present throughout the epidermis, can contribute to black/brown (eumelanin) or yellow/red colors (pheomelanin). The reddish shades can also be enhanced by hemoglobin stationed in the dermal microvascular network of the dermis.

On the other hand, carotenoids, organic pigments produced by fruits, vegetables, and algae, substantially contribute to the yellowish skin appearance. ⁴ They can be found throughout all skin layers, primarily to protect against the deleterious effects of sunlight‐induced oxidation. ⁵ The major carotenoids in human skin are lycopene, carotenes (α, β, γ, and δ), β‐cryptoxanthin, lutein, and zeaxanthin. ⁶ Despite their essential role in humans, carotenoids are not found in wild canids like wolves, foxes, and hyenas. Researchers accordingly concluded that canids are poor carotenoid accumulators despite having a diet with moderate levels of carotenoids. ⁶ , ⁷

Skin colorimetric data are normally presented in the CIELAB (Commission Internationale de l'Éclairage/International Commission on Illumination's L*a*b*‐based) color space, which consists of three coordinates: L* (lightness), a* (green‐red), and b* (blue‐yellow). The CIELAB concept was intended to be a perceptually uniform color space, where measured color differences correspond to the color changes perceived by the human eye. The measurements are normally accomplished using state‐of‐the‐art colorimeters like the CR‐400 Chroma Meter (Konica Minolta, Tokyo, Japan) ⁸ , ⁹ with a selling price of US $10 000. Alternatively, skin color can be estimated by imaging systems based on regular ¹⁰ , ¹¹ or smartphone cameras ¹² , ¹³ with the CIE76/CIE94 color errors between 2.0 and 5.9, the range of the human eye's capability to spot a color difference (i.e., just noticeable difference, JND).

The native human skin exhibited a CIELAB lightness (L*) between 30 and 75 (Table 1). However, studies have shown that skin pigmentation decreases with age and that females tend to have paler skin than males. ¹ Furthermore, the typical human skin is slightly red (an a* around 10) and yellow (a b* around 17). ² Many factors affect the skin's colorimetric results. ¹⁴ Seasons during which studies are performed have probably the most significant impact since ultraviolet (UV) light increases the melanin skin concentration during tanning. ¹⁵ On the other hand, sun exposure can decrease the amount of carotenoids like the reddish lycopene. ⁴ Other important colorimetric factors include the ambient and skin temperatures, the ambient light, contact pressure, and instrument variability.

TABLE 1.

Human native skin color in the CIELAB color space

Datasets	L*	a*	b*	Settings
Ref 2 : CA (187), EA (202), KU (145), SA (426)	28.8–73.3	1.1–19.8	6.3–24.7	D65 (2°)
Ref 16 : CA (13)	33.2–71.1	3.8–17.2	∼14.5–16.8	unknown
Ref 17 : BA (10), CA (10), EA (8), LA (10), SA (10)	∼33.3–71.4	∼4.3–15.8	∼8.8–21.6	unknown
Ref 18 : BA (9), CA (55), EA (9), SA (7)	∼31.4–75.5	∼3.6–11.8	/	D65 (10°)
Ref 19 : BA (101), CA (546), EA (1146), LA (172), SA (216)	∼30.3–72.8	/	∼4.7–25.7	D65 (10°)

Abbreviations: BA, Black African/American; CA, Caucasians; EA, East Asians: Chinese, Japanese; KU, Kurdish; LA, Hispanic/Latin; SA, South Asian: Indian, Thai,

∼Denotes an approximate color range estimation from the published figures.

Studying the genetics of animal skin and hair colors is the most popular for breeding purposes. ²⁰ In farmed animals, coat features are mostly related to economics. ²¹ On the other hand, the coats of horses have been investigated throughout human history since uniquely colored horses (especially stallions) represented a political, financial, or religious power. ²² Even today, the coat color affects the personality characteristics people attribute to dogs. ²³ Additionally, dermatoscope‐based colorimetry has been recently reported to reliably estimate the skin erythema (redness) in allergic dogs. ²⁴ , ²⁵ Since erythema is one of the most important clinical signs during allergic flares, objective skin color estimations could lead to more reliable patient monitoring and, consequently, to better clinical outcomes. ²⁶

Unlike the well‐studied human skin, the color of native canine skin is unknown. Herein, we studied the skin color of healthy dogs using a calibrated optical system with smartphones and a dermatoscope. The colorimetric data was retrieved in the CIELAB color space, and the typical ranges in each color coordinate were compared to those of human skin. Our study presents the first dataset on canine skin color, which could be used for new diagnostic and cosmetic applications in veterinary dermatology.

2. MATERIALS AND METHODS

We measured skin color in 140 healthy dogs who had visited a veterinary clinic for preventive procedures (vaccination and neutering). Two investigators performed colorimetric measurements on 32 dogs with their individual smartphones at the beginning of surgery, shortly after general anesthesia was induced. During regular physical examinations, we additionally enrolled calm and cooperative dogs (n = 108) who were not bothered by the colorimetric procedure. The skin color was measured in the inguinal region, close to the less hairy fold between the hind limb and body. Both investigators ensured that the measurements were taken before any skin manipulation like cleaning or clipping. If a dog exhibited spots of different skin colors, the investigators took a separate measurement for each of the spots. To determine the intra‐observer reliability, two consequent measurements were taken at each area.

The skin color was determined with the smartphone‐ and dermatoscope‐based optical system, described in our previous studies. ¹³ , ²⁵ Briefly, the system alternately included two smartphones (P20 Pro and P30 Pro, Huawei, Shenzhen, China) with an attached dermatoscope DermLite DL1 basic (3Gen, Inc., San Juan Capistrano, CA, USA), which provided a circular 3.0 cm² sampling area. The acquired RGB data were transformed to the CIELAB color space based on the model trained on the ColorChecker Classic (X‐rite, Grand Rapids, MI, USA). First, the brightest (white) ColorChecker patch served for image normalization. Second, we built a linear regression model between an unknown device‐dependent RGB and a device‐independent CIELAB color space:

$$LAB=a_0+a_{1}R+a_{2}G+a_{3}B,\left\{L^{\ast },a^{\ast },b^{\ast }\right\}\in LAB$$ (1)

where L* (lightness), a* (green‐red), and b* (blue‐yellow) are coordinates of the CIELAB color space, and a₀₋₃ are regression coefficients, estimated with the function fitlm (Matlab, R2016a, MathWorks, Natick, MA, USA). Since most of the studies listed results for the D65 illuminant (10° observer), our initial data (D50 illuminant, 2° observer) were converted in four steps:

$$LA{B}_{D50,2^{\circ }}{\to }^{1}XY{Z}_{D50,2^{\circ }}{\to }^{2}XY{Z}_{D65,2^{\circ }}{\to }^{3}XY{Z}_{D65,{10}^{\circ }}{\to }^{4}LA{B}_{D65,{10}^{\circ }},$$

where LAB and XYZ are the CIELAB and CIE 1931 XYZ color spaces. The transformations between both color spaces (conversions 1 and 4) were based on the functions lab2xyz and xyz2lab (Matlab R2016a). The chromatic adaptation (conversion 2) utilized the Bradford transformation matrix: [0.9555766, −0.0230393, 0.0631636; −0.0282895, 1.0099416, 0.0210077; 0.0122982, −0.0204830, 1.3299098]. Finally, the observer angle (conversion 3) was converted according to the ratio between both illuminants’ (D50, D65) white points (X = 95.04854, 94.80967; Z = 108.88300, 107.30400).

The final results were compared to two human skin datasets ² , ¹⁷ based on the Wilcoxon rank‐sum test. The Administration of the Republic of Slovenia for Food Safety, Veterinary Sector and Plant Protection approved our study under the reference number U3440‐187/2020/7.

3. RESULTS AND DISCUSSION

The mean age of the 140 involved animals was 4.1 years (age range between 2 months and 18 years). The most common breeds were Yorkshire Terrier (n = 17), Border Collie (6), Labrador Retriever (5), Chihuahua, French Bulldog, Jack Russell Terrier, and Pekinese (4); there were 31 mixed‐breed dogs. We excluded 39 animals from further colorimetric analysis since their skin images exhibited erythema (n = 6), too much hair (4), or lack of quality (29, e.g., out of focus, missing normalization, ambient light presence, etc.). Together, 134 inguinal skin color measurements were performed in the remaining 101 animals. To rule out any effect of the anesthesia on the skin color following peripheral vasoconstriction, we initially compared the values between patients with anesthesia (25 animals, 33 measurements) versus those in which the measurements were done while being conscious (76 animals, 101 measurements). As there were no significant differences (p = 0.35, 0.06, 0.68 for L*, a*, b*, respectively), all dogs were grouped together for the reminder of the analyses. In this study, we opted for healthy animals, routinely presented to veterinary clinics for preventive procedures such as neutering and vaccination. This selection criteria obviously resulted in a generally young group of dogs of various breeds. It is possible that dogs of different ages, breeds, or disease ranges might exhibit more diverse skin colors; but this will need to be addressed in future studies.

The canine skin lightness (L*) ranged from 28.5 to 78.3 (Table 2, Figure 1), which is slightly broader than the range in human skin (L* = 28.8–73.3). ² However, the Wilcoxon rank‐sum test showed no significant difference between both data sets (p = 0.99).

TABLE 2.

Canine skin color in the CIELAB color space. Separately for patients under general anesthesia or being conscious, medians are listed. The colorimetric errors (i.e., ΔE, CIE94 differences) with reference intervals (†, 2.5‐97.5 percentile) are calculated against the ColorChecker (overall and for seperate CIELAB coordinates) or a repeated measurement

	Mean	Median	[Full range]	Anesthesia	Awake	Smartphone error		Intra‐rater error
L*	57.9	60.5	[28.5‐78.3]	60.7	60.5		2.4
a*	6.1	5.6	[2.5‐13.7]	6.4	5.5	6.4 [2.2‐13.2]†	4.8	1.3 [0.2‐8.1]†
b*	11.9	12.3	[6.1‐17.5]	12.9	12.2		1.7

FIGURE 1.

Color of canine and human skin color in the CIELAB color space: (A) L*, (B) a*, and b*. D1 and D2 denote the Xiao's ² and Alaluf's ¹⁷ colorimetric datasets. p‐Values (* representing a significance) were based on the Wilcoxon rank‐sum test. Next to the canine boxplots, error bars depict the expected measurement color difference against the color calibration target (ColorChecker Classic)

Canine skin proved to be less red (median a* = 5.6 vs. 9.5, p < 0.01, Δa* = 3.9) and yellow (median b* = 12.3 vs 16.6, p < 0.01, Δb* = 4.3) than human skin (Figures 1 and 2). However, the detailed analysis of the color errors (ΔE, CIE94) against the color calibration standard (ColorChecker) showed that our optical system underestimated results in both color coordinates (averages of 4.8 and 1.7 for Δa* and Δb*, respectively). Since colorimetric errors for the a* coordinate (reddish color) are in the range of the reported skin color difference, we cannot be certain that canine skin is truly less red than human skin.

FIGURE 2.

Canine and human ² skin color in the chromatic a*b* plane (a scatter plot with marginal histograms). For display purposes only, the human dataset was downsampled by a factor of six

In contrast with redness, the canine skin exhibited a significantly lower yellowness (b*). The reason for this difference could be, as reported in the introduction, that domestic dogs cannot accumulate typically yellowish carotenoids like wild canids. In humans, some exogenous skin chromophores (like lycopene) additionally contribute to red skin shades, leading to a possible difference in red color as well. However, the difference in carotenoid‐associated redness between canine and human skin could be compensated by a hemoglobin‐rich dermal microvascular network, which seems to be closer to the surface in dogs due to their thinner epidermis than that of humans (three to four cell layers, 20–30 μm ²⁷ , ²⁸ vs. 10 cell layers, 80–90 μm in humans ²⁹ ).

The L*b* plane (Figure 3) is a popular plot used to display the typical human skin color in different geographical areas; it forms a crescent‐shaped area showing the trend between these two coordinates. In the dog, however, we did not detect any dependence between L* and b* (Spearman correlation coefficient of −0.08, linear fitting: b* = k₁ ×L* + k₂ , k₁  = −0.003, k₂  = 12.08). The canine individual typology angle (°ITA = arctan([L*−50]/b*)) ranged from −62.9° to 69.5°, which covers all the °ITA‐based human skin color groups (very light, light, intermediate, tan, brown, and dark ¹⁹ ).

FIGURE 3.

Canine (orange) and crescent‐shaped human (blue, Ly et al. ⁸ ) skin color in the L*b* plane. Bluish and violet areas (Del Bino and Bernerd ¹⁹ ) mark typical skin color in different geographical areas. BA, Black African; CA, Caucasians; EA, East Asians: Japanese; SA, South Asian: Indian

Our optical systems typically achieved color differences of around 6.4 (Table 2), which is comparable to those of other camera‐ or smartphone‐based imaging systems. ¹⁰ , ¹¹ , ¹³ Repetitive measurements on the same skin spot showed a high intra‐rater reliability. The mean CIE94 color difference was less than 2.0, which is below the visible detection threshold (JND).

4. CONCLUSIONS

In this paper, we report the first dataset of the native canine skin color in the CIELAB color space. The results showed that the lightness L* (and perhaps the skin redness a*) matched the range of human skin colors. Additional studies, which incorporate a standard state‐of‐the‐art colorimeter, are needed to verify any difference in skin redness between the two species. In contrast, canine skin turned out to be significantly less yellow (lower b*) than human skin. Furthermore, we did not find any dependence between L*and b*. Better knowing the canine native skin color range could be useful for diagnostic and cosmetic applications in veterinary dermatology.

CONFLICT OF INTEREST

The authors have declared no conflict of interest.

Cugmas B, Štruc E, Kovče U, Lužar K, Olivry T. Evaluation of native canine skin color by smartphone‐based dermatoscopy. Skin Res Technol. 2022;28:299–304. 10.1111/srt.13130