Introduction

Facial skin laxity is the result of a combination of factors. First, with age, collagen and elastin fibers in the skin gradually decrease, leading to a loss of skin elasticity, which in turn triggers skin laxity1,2,3. In addition, increased exposure to ultraviolet (UV) rays accelerates facial wrinkles and sagging4. Other external factors, such as gravity and environmental pollution, can also negatively affect facial skin5,6. For Asians, the structural features of the face and the physiological properties of the skin determine the uniqueness of the signs of aging. For example, loss of volume in specific areas of the face may be an important factor contributing to their skin laxity7.

Facial skin laxity presents itself in multiple forms, typically characterized by under-eye bags, drooping eyelids, the deepening of facial lines, the emergence of perioral lines, and a blurred jawline definition4,8,9,10,11,12,13,14. While the specific manifestations can differ from person to person, collectively they contribute to a visage of aging, potentially impacting an individual’s appearance and self-esteem15,16,17,18,19. Research indicates that individuals in Asian countries are particularly attentive to skin firmness and are more likely to seek interventions to combat aging signs at an earlier stage compared to their Western counterparts20. In addition, aesthetic standards differ due to cultural backgrounds. East Asian cultures may place more emphasis on the firmness and smoothness of the overall facial contours, while Western cultures may emphasize the three-dimensionality of facial features and natural aging. In East Asia, where lighter nasolabial folds are often considered a sign of youth21, wrinkles and aging in the mid-face region can cause the perceived age of the skin to differ from its actual age22. Therefore, East Asians may take measures to reduce the depth of nasolabial folds by injecting fillers or undergoing laser treatments23,24. In the West, people may be more inclined to accept the natural presence of nasolabial folds as part of facial expression rather than a mere sign of aging25. Hence, more race-specific factors need to be considered when dealing with Asian facial aesthetics16,26,27,28.

Current research evaluating facial skin laxity can be broadly classified into three categories: clinical assessment, imaging analysis, and the measurement of objective skin physiological parameters10,12,29,30,31,32,33,34. Clinical assessment serves as the fundamental and widely employed method for evaluating facial skin laxity. It primarily depends on the professional expertise and observational skills of the assessor, who typically employs a grading system ranging from grade I (mild) to grade IV (severe) to characterize the severity of skin laxity35. In addition to this grading system, various questionnaires and scales have been developed for clinical use36, including the validated Fitzpatrick Wrinkle, Fold, and Tissue Laxity Scale (FWFTLS)37, the Facial and Neck Laxity Grading Scale (FLR)38, and several others. Clinical assessment methods, while more intuitive, are inherently subjective and can vary between assessors, which may compromise their consistency and reliability. In contrast, imaging analysis and the measurement of objective skin physiological parameters offer a more objective and precise approach to evaluation. Common imaging techniques include ultrasound imaging and 3D imaging. Ultrasound imaging provides detailed visualization of structural changes within the skin and underlying tissues, such as the loosening of collagen fibers and the distribution of adipose tissue33,39. 3D imaging utilizes sophisticated software to create detailed three-dimensional representations of facial skin, facilitating easy comparison of changes over time and supplying quantitative data for analysis31. The Cutometer is a widely utilized device for assessing skin laxity by quantifying the skin’s elasticity and resilience. Furthermore, elastography, a method that employs ultrasound or magnetic resonance imaging (MRI) to measure the skin and dermis’s modulus of elasticity, offers a quantitative measure of skin laxity40. These advanced tools help mitigate subjective errors, enhancing the precision and reproducibility of skin laxity assessments. In conclusion, the various methods for assessing facial skin laxity each possess distinct attributes. Clinical assessments offer intuitive yet subjective insights, whereas imaging analyses and quantitative measurement tools deliver a more objective and accurate means of evaluation, providing robust data support.

However, these tools have not been fully and rigorously characterized or modeled, particularly within the context of the Chinese female population. For a comprehensive and precise evaluation of facial skin laxity, it is essential to consider employing a combination of assessment methods41. This study employed a comprehensive approach, integrating expert assessments, consumer perceptions, and noninvasive measurements of skin physiological parameters, to develop a mathematical model that characterizes facial skin laxity in Chinese females. This model establishes a scientific foundation for the development of personalized beauty and skin-firming strategies.

Materials and methods

Participants and facility

The participants of this study were healthy Chinese females aged 18-60 years old and had to meet the following conditions: (1) those who are troubled by facial skin laxity; (2) self-reported wrinkles or fine lines around the eyes and laxity; (3) the presence of pronounced facial wrinkles or signs of sagging was confirmed by a professional assessor; (4) those who are able to understand the trial process and sign an informed consent form.

The study complies with the Declaration of Helsinki and has been authorized by the participants’ portrait rights and informed consent forms (ICF). The study was conducted at Beijing EWISH Testing Technology Co., Ltd. in Beijing, China. All participants in the study were asked to clean their facial skin with the same cleanser in a laboratory setting and were acclimatized to constant temperature (21 ± 1°C) and humidity (50 ± 10%) for 30 min before image and data acquisition. A total of 142 participants who met the screening requirements had their facial images and skin physiological parameters collected during the study.

Image capture

Front, left, and right facial images of the participants were acquired using the VISIA-CR (Canfield Scientific Inc., Parsippany, NJ, USA), and the light sources of the acquired images consisted of standard light 1, standard light 2, parallel-polarized light, cross-polarized light, and UV light.

Noninvasive measurement of skin physiological parameters

The noninvasive physiological parameters measured in this study and their testing instruments and sites can be seen in Table 1.

Table 1 Non-invasive physiological parameters and their testing instruments and sites.
Full size table

Non-invasive physiological parameters of the cheek, underneath eye wrinkles, nasolabial fold, and marionette lines on either side of the participants were collected according to a randomized table. The test site of the cheek is the intersection of the vertical line extending downward from the middle of the eye and the horizontal line at the ala nasi, and a schematic diagram of the specific test site can be seen in Fig. 1.

Fig. 1
figure 1

Schematic diagram of the test sites.

Full size image

Expert assessment

Five experienced professional evaluators estimated apparent age from front VISIA-CR images taken under standard light 1 of 142 healthy Chinese women aged 18-60. Apparent age assessment requires a combination of facial skin pigmentation, skin tone, wrinkles, contour, and other dimensions of age-stage skin characteristics, and the assessment data were recorded in whole numbers. After the initial, independent, blinded evaluations, the panel convened to reach a consensus, and the resulting value was recorded as the definitive expert assessment.

Furthermore, the same five evaluators screened the front VISIA-CR images acquired under standard light 1 for facial skin laxity in all 142 participants to identify suitable candidates for the subsequent consumer perception study. Five evaluators were required to exclude the effects of skin pigmentation and skin tone, focusing on the degree of visibility of underneath eye wrinkles, nasolabial folds, and marionette lines on both sides of the face, and the degree of cheek depressions and jawline edge definition. Finally, 64 participants were selected for the consumer perception study. Participants should be selected to cover a wide range of facial skin laxity, and the study should be conducted within a manageable range. The 64 participants were labeled equally into two categories, “more facial skin laxity” and “less facial skin laxity,” by the subjective assessment of five professional evaluators.

Consumer perception

The study evaluated the VISIA-CR images taken under standard light 1 of 64 participants for consumer perception by a pairwise comparison test, also known as a two-alternative forced choice (2-AFC) test. 64 participants were randomly divided into eight groups, each of which consisted of four people labeled as having “more facial skin laxity” and four people labeled as having “less facial skin laxity.” To improve perceptual accuracy, we followed the Thurstonian modeling framework42,43,44 and comprehensively paired participants labeled as “more facial skin laxity” (M1, M2, M3, and M4) with participants labeled as “less facial skin laxity” (L1, L2, L3, and L4) in each 8-person group, resulting in 16 unduplicated pairs (shown in Table 2).

Table 2 Examples of participant matching programs.
Full size table

In order to increase the perceptual accuracy, one exchange of position was considered, so there are 32 pairs per group, totaling 256 pairs for 8 groups. A consumer perception atlas for facial skin laxity was created by randomly disrupting the order of each pair, the form of which can be seen in Fig. 2. Consumers will choose between 256 A and B pairs for “more facial skin laxity.”

Fig. 2
figure 2

Example of consumer perception atlas for facial skin laxity.

Full size image

Before consumer perception, participants were provided with a brief training session on the concept and manifestations of facial skin laxity, the objectives of their perception, and the data recording procedures. This study gathered insights from 11 cosmetic industry professionals regarding their understanding of facial skin laxity, which can be encapsulated as follows: Facial skin laxity denotes the diminished elasticity and firmness of the facial skin attributable to some factor, leading to visible sagging, looseness, and loss of smooth contours. This condition is characterized by specific signs such as drooping eye bags, pronounced eye wrinkles, deepening nasolabial folds, the formation of marionette lines, sagging apple muscles, cheek hollowing, and an indistinct jawline, as illustrated in Fig. 3.

Fig. 3
figure 3

Schematic diagram of consumer perception training for facial skin laxity.

Full size image

Consumers evaluated the front images of the paired VISIA-CR standard light 1 images in the 2-AFC test, selecting the “more facial skin laxity” from each pair of participants. In this study, it was only necessary to assess the side of A and B that had more facial skin laxity, not to determine whether there was a difference between A and B. Therefore, the statistical method of pairwise comparisons with a one-sided test was chosen to calculate the number of consumers required45. The minimum number of evaluators n is calculated from the normal approximation of the binomial distribution, as illustrated in Eq. (1). At least 68 evaluators are needed for the one-sided pairwise test under the conditions of α = 0.05, β = 0.5, and Pd = 20% (α: alpha-risk, false-positive rate; β: beta-risk, false-negative rate; Pd: percentage of evaluators who detect a discrepancy)46.

$$n \approx \frac{{z_{1 - \alpha } \sqrt {p_{0} 1 - p_{0} } + z_{1 - \beta } \sqrt {p_{1} 1 - p_{1} } }}{{p_{1} - p_{0} }}^{2}$$
(1)

\({p}_{0}=0.5\), the probability of correct choice when there is no significant difference;

\({p}_{1}={p}_{d}+(1-{p}_{d})\times 0.5=0.2+(1-0.2)\times 0.5=0.6\), the probability of correct choice when there is significant difference; \({z}_{1-\alpha }=1.64\) (α = 0.05); \({z}_{1-\beta }=0\) (β = 0.5); n ≈ 67.24.

In this study, a total of 72 participants were recruited to partake in consumer perception assessments, comprising 36 individuals with professional backgrounds in cosmetics and 36 without such expertise.

In summary, the decisions made by the 72 participants reflect the outcomes of their consumer perceptions. The study evaluated a total of 64 participant facial images, which encompassed 256 pairwise comparison events, ultimately yielding 18,432 responses in total.

Quantification of consumer perception results

The analysis of consumer responses involved calculating the odds ratio for selecting T, which represents “more facial skin laxity,” in each pair comparison. Subsequently, the logit function was applied to these odds ratios (denoted as Logit[T]), as illustrated in Eq. (2).

$${\text{Logit}}\left( T \right) = \ln \left( {\frac{{P_{T} }}{{1 - P_{T} }}} \right)$$
(2)

where p denotes the binomial probability and \(\frac{{P_{T} }}{{1 - P_{T} }}\) signifies the odds ratio of selecting T. This method facilitates the transformation of consumer responses, which are in binary format, into a dataset that follows a normal distribution. This conversion allows for the mapping of the probability of consumer choices across a continuum from negative infinity to positive infinity47. Out of the 18,432 responses collected in the pairwise comparison study, the normal order and reversed order responses were averaged to generate 128 unique high-low pairs across the 64 study participants.

Modelling

Correlations between expert apparent age assessment results, consumer perception differences, and the combined effect of noninvasive skin physiological parameters were established by the support vector machine (SVM) regression model. Consumer-perceived results demonstrate selection for “more facial skin laxity” in 128 unduplicated pairs. Quantifying consumer perception through Logit(T).

The Bradley-Terry model, a probabilistic model that allows for combining and estimating the results of pairwise comparison experiments44, was used to convert consumer-perceived Logit(T) results of facial skin laxity for 64 participants into a latent facial skin laxity score. In accordance with the Bradley-Terry scoring matrix, we integrated the raw Logit(T) scores obtained directly from consumers’ perceptions and applied a sigmoid function to the entire set of Logit(T) results. By inputting the results into the Bradley-Terry model, we derived a Bradley-Terry score for facial skin laxity among 64 participants. Subsequently, a linear relationship between the degree of facial skin laxity score and expert assessment results and objective parameters was established. See Fig. 4.

Fig. 4
figure 4

Technical roadmap for modeling the FSLI.

Full size image

Results

Distribution of noninvasive skin physiological parameters

A comprehensive raw dataset encompassing facial skin parameters was gathered from 142 participants, comprising a total of 24 parameters. This dataset included expert assessments of apparent age along with 23 noninvasive physiological skin measurements, as shown in Table 3. The Shapiro-Wilk normality test showed that some of the parameters were normally distributed and some were not normally distributed.

Table 3 Distribution properties of acquisition parameters.
Full size table

Of the 24 parameters collected for the study, 2 parameters (skin elasticity F3 and R5 values) were excluded from the modeling process due to their direct correlation with skin elasticity R2 and F4 values, as shown in Fig. 5. The remaining 22 parameters were considered independent (variance inflation factor < 5) and represent skin attributes such as moisture, wrinkles, elasticity, and collagen.

Fig. 5
figure 5

Acquisition parameter correlation.

Full size image

Consumer perception response

The results from the 256 pairs that contained swapped positions were integrated, yielding 128 unique pairs of consumer perceptions. A binomial test was conducted, and the results indicated that the consumer perceptions were significantly different across all pairs (p < 0.05, α = 0.05, two-tailed). This suggests that the consumer could significantly perceive the difference between 128 pairs of non-repeated images and choose the one with “more facial skin laxity” among them. The pairwise combinations used for consumer perceptions were grouped by expert assessment, therefore, Cohen’s kappa was used to analyze the consistency of the data from the 72 consumer assessments with the expert-assessed “standard answers.” It was found that the κ-mean of 72 consumers was 0.929 with a 95% confidence interval of [0.913, 0.945]. The κ-mean of 36 consumers with a professional background in cosmetics was 0.932 with a 95% confidence interval of [0.909, 0.956]. The κ-mean of 36 consumers without a professional background in cosmetics was 0.926 with a 95% confidence interval of [0.903, 0.949], indicating that there was a high degree of consistency between consumer perceptions and expert evaluations. Consumers with a professional background in cosmetics had a higher concordance with expert assessments compared to those without a professional background in cosmetics, but there was no significant difference between the two groups in terms of perceptions of facial skin laxity (p = 0.29 > 0.05, Mann-Whitney U test, α = 0.05, two-tailed).

The perceived outcomes for 128 unique pairs of participant atlases were analyzed and quantified utilizing the Logit(T) model. Participants exhibiting more pronounced differences in perceivable facial skin laxity corresponded to higher absolute |Logit(T)| values, with the reverse also being true. In this study, certain pairs demonstrated markedly greater differences in facial skin laxity, leading to instances where every consumer uniformly selected one face over another within those pairs. Consequently, this resulted in a slight skew in the distribution of Logit(T) values towards the higher end of the scale’s ratings.

Modelling

The Bradley-Terry score was logarithmically transformed and named the Facial Skin Laxity Index (FSLI). The model was developed using SVM regression in R, establishing a relationship between the FSLI and the apparent age assessment results, as well as the noninvasive skin physiological parameters of 64 participants. Out of the 22 parameters considered in the modeling process, 8 were ultimately selected to ensure the model’s accuracy. The outcomes of the modeling are presented in Table 4.

Table 4 R modeling results.
Full size table

R2: 0.811, the coefficient of determination, measures a model’s capacity to explain the variance in the target variable; the closer the value is to 1, the better the model fits the data. RMSE: 0.0518, the root mean square error, represents the average magnitude of the errors between predicted and actual values, indicating the degree of deviation; the closer the value is to 1, the higher the prediction accuracy of the model. Higher values of the FSLI represent a higher degree of facial skin laxity.

Correlation analysis was done between the FSLI values predicted by the model and the actual values, and the correlation coefficient between the actual values and the predicted values was 0.433, p < 0.001, which showed that there was a significant correlation between the actual values and the predicted values, indicating that the model was well fitted. Fig. 6 shows the predicted vs. reality FSLI scores (Fig. 6a) and residual plot (Fig. 6b) of the FSLI mode.

Fig. 6
figure 6

Model performance. (a) Predicted vs. reality FSLI scores; (b) Residual plot.

Full size image

Validation of FSLI model

The FSLI model described above was applied to measure facial skin laxity in 78 Chinese female who were not previously involved in modeling among the 142 participants in this study, and FSLI scores were calculated. Fig. 7 shows the histogram (Fig. 7a) and Q-Q plot (Fig. 7b) of the measured FSLI values, However, the distribution of FSLI in this population is non-normal (p = 0.004, Shapiro-Wilk normality test), which may be attributed to the relatively small sample size. The mean of this distribution is 0.3961, the median is 0.3950, and the standard deviation is 0.00436.

Fig. 7
figure 7

Distribution of facial skin laxity index (FSLI) in 78 Chinese female. (a) Histogram; (b) Q-Q plot.

Full size image

Application of FSLI model

FSLI was applied to evaluate the efficacy of oral cosmetic products combined with topical cosmetic products to improve the skin. In this example, we used parametric indicators from a previous clinical study. The study had been registered with the Chinese Clinical Trial Registry (registration No. ChiCTR2400085413, 07/04/2024) and had passed the ethical review by the Shanghai Ethics Committee for Clinical Research (approval No. SECCR/2024-46-01, 06/06/2024). In that study, 15 Chinese female participants used oral and topical beauty products consecutively to quantify the changes in facial skin laxity through metrics collection and data analysis. Higher values of the FSLI represent a higher degree of facial skin laxity. As can be seen in Fig. 8, FSLI significantly decreased after 12 weeks of using the test products compared to baseline (p = 0.007, normal distribution, paired sample t-test, α = 0.05, two-tailed).

Fig. 8
figure 8

Change in 12-week FSLI for oral and topical beauty products.

Full size image

As shown in Fig. 9, analysis of non-invasive skin physiological parameters associated with the FSLI model showed that ET, R2, Q1, and CC2 were significantly improved at 12 weeks (p ≤ 0.001, normal distribution, paired sample t-test, α = 0.05, two-tailed). UEW_V, ML_V, and NF_V were significantly decreased at 12 weeks (p = 0.001, non-normal distribution, Wilcoxon signed-rank test, α = 0.05, two-tailed). F4 improvement, although not significant (p = 0.064 > 0.05, normal distribution, paired sample t-test, α = 0.05, two-tailed), still tended to decrease. All of the above changes indicate that the skin condition has become better and tends to tighten.

Fig. 9
figure 9

Change in 12-week non-invasive physiological parameters for oral and topical beauty products. *p < 0.05, **p < 0.01, ***p < 0.001, α = 0.05, two-tailed. (a), (e), (f), (h) Increases in ET, R2, Q1, and CC2 indicate that the skin condition has become better and tends to tighten; (b–d), (g) Decreases in UEW_V, ML_V, NF_V, and F4 indicate that the skin condition has become better and tends to tighten.

Full size image

Significant lightening of underneath eye wrinkles, nasolabial folds, and marionette lines at 12 weeks as shown in Fig. 10.

Fig. 10
figure 10

Significant lightening of wrinkles at 12 weeks. (a) Underneath eye wrinkles; (b) Nasolabial folds; (c) Marionette lines.

Full size image

Discussion

Facial skin laxity is one of the more recognizable and perceptible facial conditions that Asians experience in their daily lives. Understanding how it is perceived and the underlying components of this phenomenon can help the beauty and skincare industry provide consumers with effective beauty management interventions. By examining this phenomenon through expert assessment, consumer perception, and noninvasive skin physiological measurements, we can achieve a more scientifically grounded characterization of aging skin. This approach facilitates the development of a daily skincare regimen that effectively addresses skin conditions. In this study, we assessed the apparent age of the participants’ skin, learned about the perception of facial skin laxity in the Chinese population, and measured relevant noninvasive skin physiological parameters. In addition, we investigated their joint effect by means of SVM regression modeling, which provided a key tool to quantitatively characterize skin clarity. Consistent with expectations, facial sagging was negatively correlated with skin elasticity R2 values and positively correlated with skin elasticity F4 values. When assigned ordinal ratings via the Bradley-Terry model, only 57 of the 64 participants received the appropriate rating, due to the fact that seven participants’ degree of facial collapse was always judged to be either low or high when compared. Therefore, when performing regression fitting with objective parameters, only data from 57 participants were modeled, which is a small sample size. Therefore, the validation method chosen in the regression modeling process of the support vector machine was the leave-one-out method, and the final model accuracy was 0.811, indicating that the model has good fitting ability.

However, it is important to acknowledge that the data in this study were derived from a previous human efficacy evaluation program with participants who self-identified as having trouble with skin laxity and who were screened for elasticity and wrinkle grades, which may have led to a decrease in the accuracy of the model in a population with mild or extremely slight laxity. In addition, the study mainly used subjective assessment and noninvasive skin physiological parameters, which may have some subjective bias. In conclusion, further longitudinal validation of the FSLI model over time or across product categories is still needed. Histologic correlation data support, such as collagen and elastin biopsies, may also be considered in future studies.

Given that the concept of loose facial skin is influenced by the cultural background of the perceiver, the model developed in this study only reflects the perceptions of Chinese or Asian individuals who have a similar appreciation for certain skin characteristics. If interest emerges in other parts of the world, other ethnicities’ perceptions of facial skin laxity should be studied and modeled. We hope that this study will provide a methodological framework for future global studies on modeling facial skin laxity.

Conclusions

Based on the findings of this study, we conclude that facial skin laxity can be effectively characterized, quantified, and predicted using a linear model that incorporates objectively measured, noninvasive physiological parameters of the facial skin. This study contributes valuable data for both expert assessment and consumer-perceived evaluation of facial skin laxity, enabling the objective quantification of skin looseness through the non-invasive measurement of physiological skin parameters.