AI-Based Facial Analysis vs Self-Report: A Pilot Study Using Insights From Pre-Procedural Psychological Tests

Introduction

The application of artificial intelligence (AI) has grown significantly, especially in the field of medicine, opening novel opportunities for research and development, while empowering patients with tools for self-education, self-monitoring, and enhanced online healthcare services [1,2].

The demand for cosmetic and plastic surgery procedures is growing due to the desire to adapt to the rapidly changing ideals of beauty promoted in social media and television, which emphasize modern standards of beauty and images of strength, attractiveness, extroversion, and self-confidence [3]. In plastic surgery, artificial intelligence-based instruments are becoming increasingly important due to their potential to improve treatment outcomes and patient satisfaction. Some AI facial qualification systems enable the acquisition of realistic 3D facial images, facilitating an accurate assessment of surgical alternatives and possible outcomes, and enabling the patient to compare “before-and-after” images, and allowing the patient to choose the optimal course of action and reduce the chances of dissatisfaction with the result [4].

While AI has considerable potential in cosmetic surgery, its use raises critical concerns. Despite providing objective evaluations, AI systems often fall short in capturing the complexities of a patient’s self-perception and neglect psychological factors like self-esteem, personality traits, and overall life satisfaction. This disparity highlights an important knowledge gap: determining how AI assessments align with patients’ self-evaluations and the influence of psychological aspects on these interpretations [4].

This study evaluated the relationship between artificial intelligence-driven applications and self-assessments made by participants using the Self-Perception Scale, focusing on the same areas in relation to dimensions such as attractiveness, competence, calmness, joy, dominance, health, and trustworthiness, as well as to analyze the correlation of selected psychological elements such as self-esteem or life satisfaction that can influence both self-assessment and external assessments made by AI. Understanding these connections is essential for the responsible and efficient incorporation of AI into cosmetic surgery. Although AI has potential for greater objectivity and accuracy, its shortcomings highlight the importance of using a balanced strategy that includes self-assessments and psychological considerations. While AI-based assessments provide a standardized, objective analysis of facial features, self-assessment ratings are inherently subjective, shaped by internal psychological states such as self-esteem and life satisfaction. Comparing these 2 approaches reveals important discrepancies. Self-assessment ratings are often more favorable than AI evaluations and strongly correlate with psychological traits. Understanding these relationships is crucial for responsibly integrating AI into cosmetic surgery planning, ensuring that both objective analysis and the patient’s psychological profile are considered. This integrated approach not only enhances the use of AI in aesthetic medicine but also fosters its growth by prioritizing a human-centered perspective.

Material and Methods

PATIENT SELECTION:

The size of the study population was calculated for a confidence level of 90%, a margin of error of 10, a population proportion of 20, and a population size 81 000 (the estimated number of medical students in Poland). The obtained sample size was 44 or more. Healthy volunteers aged 18- and 26-years were recruited for the control group. The inclusion criteria were age 18–26 years, a good general state of health, and no known psychological or dermatological conditions. No additional demographic or educational criteria were applied. Participants were recruited via Facebook by posting an invitation on groups associated with the Medical University of Lublin. The invitation was directed to students of the university, and no targeting or selection methods were employed beyond this criterion.

The study procedure involved taking facial photographs and completing prepared psychological tests. The photography session required approximately 3 minutes per participant, and completing the questionnaires took about 12 minutes, totaling approximately 15 minutes per participant.

In the questionnaire, participants were asked to provide basic demographic information such as gender, the population size of their place of residence, educational level, and field of study. No sensitive data were collected; in this context, “sensitive data” refers to personal identification numbers (eg, PESEL number), information revealing racial or ethnic origin, political opinions, religious or philosophical beliefs, trade union membership, genetic data, biometric data used for identification purposes, or data concerning health, sexuality, or sexual orientation. Emphasizing the absence of collecting sensitive data is due to the study’s focus on the visual assessment of participants without considering their lifestyle or other private aspects.

AI-BASED TOOL:

To enable coherent assessment, an artificial intelligence-based tool was used. The program used during evaluation simulates human first impressions and assess 7 areas: attractiveness, competence, restedness, happiness, dominance, health, trustworthiness. Initially, the patient’s face is photographed, then the results are presented as a chart and on a scale.

The results of assessing participants using the AI method are presented in a graphical depiction, with a grid divided into 14 equal parts. This form of precisely measuring the graphical outputs was provided by the application, proposed by the originators of the tool, and is a form of obtaining the results of the study using the AI method. Therefore, in the work presented here, no validation of the scale was carried out. Since this was a pilot study, we decided to use the existing form of obtaining results. The main argument in favor of this was the division of the scale into equal parts and an equal scale (14-point ranges) for each of the 7 dimensions included in the study. The result of testing a person with this tool was therefore a graphical depiction, in which the outcome of the study on a given dimension was overlaid on a 14-compartment graphical chart. These results were read out by counting down the intervals on the graph that the surveyed person obtained on a given scale. The visual scales were divided into 14 equal segments using measurement tools, assigning numerical values to each segment to quantify the AI assessment. This approach ensured a standardized scoring system across all participants, enhancing the reliability of the measurements.

The AI tool used was FaceReality Application (https://face-reality.com). All participants were assessed using the same version of the application to maintain consistency throughout the study. The application uses neuronal networks and learning algorithms to correlate facial features. The data were validated by the Medical University in Vienna.

In our protocol we started from the psychological tests, described below, and just after filling in all the questions we took a picture with the application. We wanted to prevent the participants from being prejudiced by the results of AI evaluation. The application is used on a tablet that has access to the network. The same device was used for all participants to ensure consistency. Photographs were taken against a plain white background in natural lighting conditions. Participants were instructed to remove makeup and jewelry in advance, maintain a neutral facial expression, face the camera directly, and position themselves at a standardized distance from the camera. After a photograph was taken, the application created heat maps according to the algorithm to visualize the facial features on the scale (Figure 1A, 1B).

Images were processed directly by the application on the device. Photographs were then securely uploaded to a private Google Drive account, which was password-protected and accessible only to authorized research personnel. Data were stored securely, with access restricted to authorized research personnel only, in compliance with data protection regulations.

Ethics approval for the use of AI and facial imaging was obtained from the appropriate ethics committee. Participants provided informed consent specifically for facial imaging and AI-based analysis, acknowledging the use of their images for research purposes under strict confidentiality protocols.

PSYCHOLOGICAL TESTS:

To assess psychological correlates relevant to the study’s objectives, 4 scales were employed: the Ten-Item Personality Inventory Polish version (TIPI-PL) [5], the Rosenberg Self-Esteem Scale (SES) [6], the Satisfaction With Life Scale (SWLS) [7], and the Self-Perception Scale (SPS), which was developed specifically for this study.

The TIPI-PL [5] is a concise instrument consisting of 10 items designed to measure the 5 major dimensions of personality according to the Big Five model: Extraversion (E), Agreeableness (A), Conscientiousness (C), Openness to Experience (O), and Emotional Stability (N). The Polish adaptation by Sorokowska et al was utilized. Participants rate statements on a 7-point Likert scale from 1 (“strongly disagree”) to 7 (“strongly agree”). Previous research reported internal consistency coefficients (Cronbach’s alpha) for the subscales as follows: O (α=0.45), C (α=0.80), E (α=0.74), A (α=0.54), and N (α=0.83) [8], with test-retest reliability correlations ranging from 0.66 to 0.74.

The Rosenberg Self-Esteem Scale (SES) measures global self-esteem as a relatively stable characteristic. It comprises 10 items with responses on a 4-point scale, yielding total scores between 10 and 40 points [9]. The Polish version demonstrated reliability coefficients (Cronbach’s alpha) between 0.81 and 0.83 in various samples.

The Satisfaction With Life Scale (SWLS) [7] evaluates overall life satisfaction through 5 items rated on a 7-point scale. The Polish version showed satisfactory reliability with a Cronbach’s alpha of 0.81 in a sample of 371 individuals, comparable to the original version’s α=0.87 [10].

The Self-Perception Scale (SPS) is an original instrument developed for this study to measure participants’ self-assessments in 7 domains corresponding to the dimensions evaluated by the AI-based tool: attractiveness, competence, restedness, happiness, dominance, health, and trustworthiness. Participants rate each dimension on a 7-point scale from 1 (“strongly disagree”) to 7 (“strongly agree”). The SPS was designed to facilitate direct comparison between participants’ self-perceptions and AI-generated assessments.

The SPS demonstrated acceptable reliability, with a Cronbach’s alpha of 0.74 (P<0.01, n=100). It showed statistically significant correlations with established psychological measures: the Body Esteem Scale for Adolescents and Adults (BESAA) by Mendelson et al (r=0.541, P<0.01), the Perceived Stress Scale (PSS) by Cohen et al (r=−0.480, P<0.01), and the Psychological Well-Being Scales by Ryff (r=0.577, P<0.01). In the pilot study sample (n=47), the SPS had a Cronbach’s alpha coefficient of 0.66 (P<0.01).

The research conducted with the SPS scale constructed for this study showed correlations of this scale with existing scales that investigate similar psychological constructs. This preliminary step in the scale’s construction provided the rationale for taking further steps toward validating this scale. In the validation process, the SPS scale as well as other scales to determine external consistency were surveyed among more than 100 respondents. Psychometric properties were calculated for the presented method. Cronbach’s alpha coefficient for each scale is as follows: Attractive α=0.77, Competent α=0.874, Rested α=0.778, Happy α=0.934, Dominant α=0.685, Healthy α=0.892, Trustworthy α=0.834. Also, in the study for validation purposes on a larger number of participants, there were numerous correlations of individual scales of the SPS method with scales within existing tools, including appearance, attribution, self-esteem, and personality dimensions.

The selection of these scales was intended to capture key psychological constructs that may influence both self-perception and external assessments by AI. The TIPI-PL assesses fundamental personality traits, the SES measures self-esteem, and the SWLS evaluates life satisfaction – all factors potentially affecting self-perception and how individuals are perceived by others or AI algorithms. The SPS allows for a direct comparison of self-assessed attributes with those generated by the AI tool in corresponding domains. References to Polish validation studies support the use of these scales in the current research context.

ETHICS STATEMENT:

The principles outlined in the Declaration of Helsinki were respected. The Medical University of Lublin Ethics Committee accepted both the study’s protocol and the subjects’ participation (reference number: KE-0254/17/01/2024). All participants were provided with written information about the study and its procedures. Written informed consent was obtained from each participant, including consent for facial imaging and analysis using the AI-based tool.

To ensure the privacy and confidentiality of participants’ data, personal information was anonymized. Only the researchers had access to the data. Data, including facial images, were stored on a private Google Drive account.

All participants were provided with written information detailing the study’s objectives, procedures, and data management practices. Informed consent was obtained from each participant before their inclusion, explicitly covering facial imaging and AI-based analysis. To ensure privacy and confidentiality, all personal data were anonymized. Only authorized research personnel had access to the stored data, which included facial images securely maintained on a private, password-protected Google Drive account.

The study collected only non-sensitive demographic information such as age, gender, and educational background. No data related to race, ethnicity, health, or other protected categories were recorded. These measures were implemented to uphold ethical standards and safeguard participant confidentiality while ensuring compliance with data protection regulations.

The authors declare they have no conflicts of interest. All figures presented in this manuscript are original, created by the authors, and have not been previously published in whole or in part.

STATISTICAL ANALYSIS:

Statistical analyses were performed using Statistica StatSoft Polska version 13.1 with default settings. Statistical significance was defined as P values less than 0.05. Descriptive statistics included the mean, standard deviation, range of values, and percentages. Cases with missing data were excluded from the analyses to ensure accuracy and consistency of the results. The questionnaires included in the study were fully completed and had no missing data.

For univariate analysis, the Shapiro-Wilk test was employed to verify the normality of the distribution of quantitative variables. Homogeneity of variances for data meeting the normal distribution criteria was investigated using Levene’s and Brown-Forsythe tests. Variables with homogeneous variances and normal distributions were compared using the independent-samples t test, which is suitable for assessing differences between 2 groups under these conditions.

The statistical analyses performed included a comparison between a group of male and female respondents in terms of the results obtained in the AI survey, as well as results of the psychological tests conducted.

Even though the resulting distribution has the characteristics of a normal distribution, for further analysis it was decided to use the Mann-Whitney U test, due to the small size of the group of subjects, to better reflect the profile of the group presented. The Mann-Whitney U test was used particularly to assess the relationship between self-assessment scores on the SPS scale and AI evaluations.

Friedman’s ANOVA was used for analyzing repeated measures of multivariate variables to detect differences across multiple related samples. Pearson’s correlation coefficient was used to examine the relationships between continuous variables included in the study.

Results

A total of 50 healthy volunteers were enrolled in the study. Three participants were excluded from the final statistical analysis due to incomplete participation in the study process, specifically not fully completing the questionnaires. Consequently, data from 47 participants were included in the analysis.

Respondents of the final group consisted mainly of females (70%, n=33). The mean age of the participants was 22 (SD2, range 18–27). Most participants were university students (89%, n=42). This group represented a homogeneous sample for the pilot study comparing AI facial analysis with self-reports presented in this paper. Figures 2A, 2B, 3A, and 3B present graphical representation of the results obtained from the AI-based application. The values of the results from the psychological tests are displayed in Table 1.

The AI-based analysis revealed significant differences with regard to gender. Female faces were rated higher in the following dimensions: attractiveness (P<0.001), restfulness (P=0.004), happiness (P<0.001), and trustworthiness (P<0.001). These findings may be related to a potential gender bias in the algorithm used by the AI tool, as well as with potential social assumptions about the evaluation of women’s faces, which could be the basis for designing an algorithm in the AI-based tool. There was a correlation between ‘Competent’ and ‘Rested’ AI results with TIPI Extraversion that approached conventional statistical significance (P=0.045), although it remained relatively weak and should be interpreted with caution.

None of the other results of the analysis of psychological status, personality types, self-esteem, and well-being reached statistical significance. The statistical methods used show the correlations between the statistically significant variables. Correlations concern personal traits, psychological measures, and self-perception assessment. Results from psychological tests as well as the SPS, constructed for this study, were included in the correlation analysis. Correlating these measurements with each other was intended to test the interrelationships with the psychological variables included in the study, while providing a rationale for their selection. This is because the correlation, which is significant at the statistical level, indicates the interrelatedness of the different areas included in the study. The occurrence of correlations between scales already validated and the new SPS scale may be a good predictor for further work on the SPS scale and an attempt to validate this method.

A moderate positive significant correlation was found between self-esteem and emotional stability (r=.523, P≤0.01), and there was a weak positive correlation with life satisfaction (r=.355, P≤0.05). This may suggest that emotionally stable individuals develop positive self-esteem, which can contribute to a better satisfaction with life. Self-esteem also had a moderate positive correlation with attractiveness (r=.401, P≤0.01), suggesting that self-esteem refers to and is based on a belief in one’s physical attractiveness.

In addition, the self-esteem variable showed a weak correlation with extraversion (r=.331, P≤0.05), perceiving oneself as competent (r=.350, P≤0.05), happy (r=.377, P≤0.01), dominant (r=.368, P≤0.05), and trustworthy (r=.350, P≤0.05).

The analysis of the correlation between personality traits and self-perception showed that extraversion has a positive correlation with attractiveness (r=.463, P≤0.01) and happiness (r=.371, P≤0.05), with may mean that more extraverted people have a more positive view of themselves in some of these areas. Conscientiousness was also correlated with attractiveness (r=. 469, P≤0.01), as well as perceiving oneself as trustworthy (r=.374, P≤0.01), which may reflect that self-disciplined individuals invest in their physical appearance and competence as well as trustworthiness.

Agreeableness has a negative correlation with perceiving oneself as dominant (r=−.426, P≤0.01), suggesting that individuals who are more agreeable tend to view themselves as less dominant, which is consistent with theoretical expectations about these personality dimensions. Also, emotional stability has a positive correlation with perceiving oneself as happy (r=.367, P≤0.05) and dominant (r=.316, P≤0.05).

The strongest correlation was between attractiveness and competence (r=658, P≤0.01), as well as with trustworthiness (r=541, P≤0.01). These strong correlations may indicate a potential halo effect, in which positive self-perception in one domain can generalize to other domains. Moreover, perceiving oneself as happy has a positive correlation with health (r=505, P≤0.01), which may indicate a relationship between psychological and physical dimensions of human functioning. There was also a strong correlation between satisfaction with life and happiness (r=.660, P≤0.01), suggesting the importance of self-perception of being happy in feeling satisfied with life.

The research results presented above indicate numerous significant intercorrelations between variables originating from the area of personality traits, variables representing psychological well-being, and the aspects related to self-perception. These relationships may be important in the process of forming self-judgements and in the tendencies young people use to self-assess in different domains. Strong correlations resulting from statistical calculations based on the surveys conducted can indicate how self-assessment of a particular aspect may relate to other areas of functioning.

The Mann-Whitney U test was used to compare AI-based tool results with self-assessment made by using the SPS scale. The analysis showed that the participants consistently rated themselves higher than the AI system on most of the dimensions: Attractive (U=507.500, Z=−4.698, P<0.001), Competent (U=165.000, Z=−7.303, P<0.001), Happy (U=656.000, Z=−3.508, P<0.001), Healthy (U=603.500, Z=−3.956, P<0.001), and Trustworthy (U=148.500, Z=−7.382, P<0.001). The only exception is for the Restedness dimension (U=756.500, Z=−2.715, P<0.05) (Table 2). It can be concluded that personal opinion consistently differed from the AI-based tool’s evaluation and was markedly higher in most cases, significantly impacting patient counseling if AI underestimates self-perceived attributes.

In relation to the aim of the study presented in the introduction, the relationships between the different variables related to one’s self-perception are outlined. In the context of the research results obtained, it seems particularly important to balance the conduct of assessments using AI-based tools with self-assessment, as there may be noticeable incongruence between the 2 types of assessment. In addition, the self-assessment process may involve a number of variables, such as personality traits, which may be difficult for AI tools to incorporate. A human-centered perspective should be prioritized, ensuring surgeons actively address algorithmic biases and fully respect patient autonomy, while allowing AI tools to play a supportive, complementary role in patient preparation for cosmetic surgeries.

Discussion

We compared the AI application with a self-assessment scale to check if there are any dependencies. In our study, the AI tool assessed faces in 7 categories: attractiveness, competence, restedness, happiness, dominance, health, and trustworthiness. The same criteria were used by the SPS scale to assess patients’ self-perception. Other psychological tests used in the study allowed us to measure personality dimensions, self-esteem, and life satisfaction, providing a broader psychological context for interpretation. By examining the interplay between AI-based impressions and these psychological measures, our intent was to ascertain whether automated, algorithm-driven ratings could offer supplementary insights in facial cosmetic assessments without supplanting established clinical evaluations. The results obtained with these tools were compared with the SPS scale to check for possible correlations between the scales. Our research revealed that, when compared to self-assessment psychological tests, the AI tool results, based on a neural network to imitate human initial impressions and judge human faces, were lower in 6 out of 7 evaluation categories. To the best of our knowledge, this is the first study to compare AI findings with psychological tests in the field of an individual’s face perception.

An additional advantage of using AI in psychology is the greater availability of psychological services to society, although the discussion should underscore the need to verify algorithmic precision and suitability for diverse clinical scenarios. Thanks to AI systems, patients can receive faster support and diagnosis. In addition, AI can support diagnoses by analyzing data from therapy sessions such as emotions, facial expressions, and body language, which can be effective in monitoring the effects of treatment. Algorithmic limitations – such as incomplete training data or errors introduced by human oversight – can compromise assessment accuracy, reminding us that subjective experiences and nuanced clinical judgments remain indispensable.

Regarding the aforementioned, there is a concern that artificial intelligence (AI) may soon replace medical professionals, particularly in specializations where technology-based programs are used. However, rather than restating our results, it bears emphasis that high technological proficiency does not eliminate the possibility of algorithmic missteps, and such oversights can lead to flawed diagnoses or misguided therapeutic suggestions. Data security and privacy, especially with large-scale medical datasets, also pose significant ethical and legal challenges. A doctor’s “soft” skills – like empathy and compassion – are irreplaceable in patient interactions, reinforcing that any automated system must be integrated thoughtfully, not merely substituted for direct clinical care [4]. Our findings illustrate that algorithmic limitations may particularly affect subtle attributes tied to emotional and social factors. AI may fail to capture complex psychosocial influences on facial expression, such as self-esteem-driven perception biases or gender-related factors influencing self-assessment, as highlighted by Tartaglia and Rollero regarding attractiveness and occupational status effects on perceived personality traits.

A person’s self-assessment seems to be very important in terms of decision-making and follow-up in the context of aesthetic medical procedures. Chen and colleagues employed the Contingencies of Self-worth Scale (scoring 1–7) and the Rosenberg Self-Esteem Scale (range 1–7) to evaluate the relationship between attitudes towards plastic surgery and self-esteem, social media use, and applications that alter pictures on social media. According to the study’s findings, using specific social media platforms and picture editors may make plastic surgery more acceptable while also having a substantial negative impact on self-esteem. This phenomenon provides a psychological explanation for why participants in our study, despite having lower AI scores, rated their own attractiveness more highly – social media contexts promote upward comparison and positive self-presentation, thereby widening the self- versus AI-assessment gap [11]. Psychological tests were also reported to evaluate the outcomes of esthetic procedures. Al Ghadeer et al used 3 scales to verify the willingness to undergo plastic surgery. The researchers employed the Body Appreciation Scale (the BAS-5-point scale is a general assessment of body satisfaction), the Rosenberg Self-Esteem Scale, and the Acceptance of Cosmetic Surgery Scale (the ACSS-7-item scale consists of 5 items that assesses the extent to which a person believes that plastic surgery will give them self-confidence and whether they would be able to undergo surgery for social reasons). As an outcome, only a small number of recipients would decide to undergo plastic surgery, while 75% reported that improvement in body appearance would be their main motivation for cosmetic procedures [12].

The multidimensional body self-relations questionnaire-appearance scales (MBSRQ-AS) and the Rosenberg Self-Esteem Scale (RSE-S) were employed prior to plastic surgery and 1 month following the procedure in Wang et al’s study. Individuals suffering from body dysmorphic disorder have become more interested in aesthetic medicine treatments, but their level of dissatisfaction with the outcome is higher than that of other patients who have shown considerable improvement in their questionnaire scores [13].

The MBSRQ-AS was also used in the study by Sarwer et al to show improvement in the level of self-esteem 6 months after cosmetic surgery [14]. However, using the same psychological tool, Bashizadeh et al did not observe significant change in postoperative self-esteem levels [15]. The RES scale was used in several studies analyzing the outcomes of cosmetic procedures [16], blepharoplasty [17], and rhinoplasty [18]. Although data published by Viana [19] revealed the opposite, studies by Alderman et al and Chowdhury demonstrated an improvement in self-esteem following the surgery.

In our study, we used TIPI-PL, Rosenberg SES Scale, SWLS, and SPS to compare them with the results of the facial assessment by the AI tool. There was a weak association between ‘Competent’ and ‘Rested’ AI results with TIPI Extraversion, with a P value close to the conventional significance threshold (P=0.045). None of the other results of the analysis of personality dimensions, self-esteem, and life satisfaction reached statistical significance. This suggests that AI assessments are modestly sensitive to overt personality traits like extraversion, but are likely insufficient to capture deeper psychological constructs or nuanced interpersonal perceptions without incorporating self-report tools. These results suggest that use of preoperative AI examination alone might not distinguish a patient’s psychological attributes.

Meeting patient expectations and evaluating facial flaws is not a novel approach. The FACE-Q is a detailed patient-reported outcome tool designed to assess main concepts and symptoms in patients undergoing both surgical and nonsurgical facial cosmetic procedures. The more than 40 independently operating scales and checklists included in this instrument allow for a comprehensive assessment of patient perspectives regarding their psychological well-being, social functioning, and satisfaction with cosmetic surgeries [20]. Also, De Maio suggested a soft-tissue filler injection algorithm based on the patient’s expectations and assessment of their facial features. The protocol is designed to improve negative facial features: tiredness, sadness, threatening expression, and sagging of the skin, as well as to improve positive attributes such as attractiveness, masculine or feminine appearance, and rejuvenation [21]. Li et al conducted a study wherein AI was employed to assess face expressions during cosmetic procedures. The methodology included an extensive evaluation of both static and dynamic facial expressions, enabling accurate measurements to be taken both before and after treatments. The program, Customized Precision Facial Assessment (CPFA), identified areas of cosmetic concern and helped physicians optimize therapies with minimal involvement by detecting micro-expressions and the active action units of facial muscles [22].

In the present era of increasing demand in facial esthetic procedures, development of a clinically meaningful instrument tailored to cosmetic surgery patients, capable of assessing psychological distress, inappropriate motivations, or expectations, should involve qualitative patient interviews and professional input, utilizing modern psychometric methods to ensure validity and reliability [23]. A narrative review by Rehman et al explored use of mental health screening tools in facial cosmetic surgery, shedding light on the prevalence of mental health conditions among candidates for such procedures. The analysis, encapsulating data from 12 articles and involving 2194 participants, revealed a high prevalence (34.1%) of potential mental health problems, with body dysmorphic disorder (BDD) emerging as the most common diagnosis. The study advocated for the integration of mental health screening as a routine part of evaluation for individuals seeking facial cosmetic surgery [24]. The conclusions drawn from the literature on how cosmetic surgery affects self-esteem differ. According to some studies, patients’ general psychological health was enhanced by aesthetic operations, such as breast augmentation, rhinoplasty, and blepharoplasty, with an emphasis on body image, self-esteem, anxiety, and depression [14,17,25], but other studies produced different results [15,26]. Numerous variables were shown to have the ability to affect psychological results, including age, gender, type of cosmetic treatment, education level, postoperative healing time, and preoperative mental state. Following surgery, patients with substantial psychological distress or previous body image issues were less likely to see improvements in their psychological state [27].

Our study did not specifically evaluate psychological disorders, yet we acknowledge the relevance of body dysmorphic disorder (BDD) as a potent confounder in interpreting both self-assessment and AI-based assessments of facial features. BDD is a psychopathology that affects up to 2.5% of the general population [28] and up to 15% of people seeking cosmetic operations [29,30]. An obsession with imagined defects in one’s appearance is the hallmark of this disorder [30,31,32]. There are regional differences in the prevalence of BDD – 8.5–35.2% in the US, 11.9% in the UK, and 14% in Brazil [33,34]. Patients with BDD frequently feel ashamed or disgusted by perceived body flaws. According to Phillips et al, BDD patients most frequently have concerns with their skin, hair, nose, and eyes. Women are more often concerned with the appearance of their breasts, while men focus on muscles [34]. In our study, we focused on facial features only. We found that the AI tool evaluated the faces of the females in our study population more favorably in the categories of attractiveness (P<0.001), restedness (P=0.004), happiness (P<0.001), and trustworthiness (P<0.001). Gender may also play a role in how satisfied a patient is with the cosmetic operation and the whole surgical experience. In overall satisfaction, male patients typically view interactions with doctors as important, but female patients place a greater emphasis on their interactions with nurses [35].

High levels of appearance-related worry are linked to medical conditions, which can cause problems in many areas of life [33]. Up to 75% of people with BDD seek cosmetic surgeries, and 66% of them undergo invasive esthetic procedures. Post-procedure data show that 16% of BDD patients report worsening symptoms of BDD and 75% report unhappiness, highlighting the importance of cautious screening in dermatological and cosmetic contexts to avoid unforeseen outcomes and additional social and clinical hazards [3]. Also, legal aspects of unsatisfied patients with BDD are often faced by the surgeons [29]. Patient satisfaction following aesthetic surgery is contingent upon both the treatment and the perioperative care [35]. The variability in postoperative satisfaction among individuals with disparate mental health baselines emphasizes the need for a unified approach that merges AI evaluations with empathetic, clinically informed decision-making.

An increase in the frequency of esthetic surgical and nonsurgical cosmetic procedures are reported globally [31,36]. Various AI applications are being employed in plastic and aesthetic surgery in practice, but their role in perioperative assessment remains incompletely examined in clinical studies. AI may be used in plastic and reconstructive surgery for a variety of purposes, such as diagnosis, anatomical landmarks for surgical planning, predicting indications for surgical reconstruction, likelihood of clinical outcomes, intraoperative decision-making support, identifying and forecasting surgical problems, outcome prediction, and postoperative evaluation [37,38]. Most AI tools are still primarily used in preclinical research. Preoperatively, a three-dimensional (3D) simulation for breast augmentation allows for the visualization of the desired aesthetic outcome based on the size of the implant. Most respondents (85–90%) in the Gupta et al study said they would definitely recommend using the 3D simulation to their friends and family after finding it to be very useful in selecting the desired breast implant [39]. BreastGAN is an example of an artificial intelligence program that was created to imitate the effects of breast augmentation using real clinical before-and-after images. The goal of training the generational adversarial network (GAN) was to mimic a surgical outcome by manipulating prior photos. When the AI-generated photos were compared to actual surgical results, the study discovered that they were similar. According to preliminary findings, the instruments appear to be a viable means of assisting patients in better recognizing and understanding possible surgical results [40]. Drawbacks of preoperative AI-based simulation include data quality assurance and standardization, privacy and ethical issues, and the requirement for interdisciplinary collaboration [37,38].

Personality traits such as conscientiousness, emotional stability, and extraversion were found to be mediated by self-esteem in shaping body esteem, which aligns with our findings showing personality traits appear to influence certain AI-assessed attributes like attractiveness and competence [41].

Notably, our study revealed that self-assessments consistently yielded higher scores than AI assessments in evaluating traits such as attractiveness and trustworthiness. Furthermore, the robust correlations observed between the Rosenberg Self-Esteem Scale and various dimensions of the SPS, including attractiveness and happiness, reaffirm the substantial impact of self-esteem on self-perception. This supports the mediational role of self-esteem suggested by Skorek et al, reinforcing its pivotal influence in both self-perceived and externally assessed physical attributes.

Attractiveness increased the perception of conscientiousness in women, which correlates with our findings showing a stronger correlation between perceived attractiveness and traits such as trustworthiness and competence, suggesting that attractiveness may have diverse implications depending on the context and gender of the assessed individual [42]. Similarly, our findings on differences between self-assessments and AI assessments in evaluating traits such as attractiveness and trustworthiness are reflected in Tartaglia and Rollero’s observations, where occupational status influenced the perception of conscientiousness but did not affect extraversion, which may indicate that some traits are more susceptible to contextual influence than others. Attractiveness exerts a stronger positive impact on personality impressions for females than for males [42]. A training data bias is also possible: commercial image sets over-represent smiling female faces, inflating AI ratings, whereas women – subjected to stricter beauty norms – often overestimate their own looks, further widening the self-versus AI discrepancy. These insights are consistent with our interpretation that self-assessments may be more positive due to self-esteem mediation, aligning with the self-esteem mediation described by Skorek et al.

A study conducted to determine the correlation between the perfectionist personality and depression as well as anxiety among patients who have undergone dental aesthetic treatment reported that the tendency to experience fear of making mistakes and uncertainty affects the higher incidence of anxiety and other symptoms before aesthetic medicine procedures [43]. Moreover, elderly patients were less likely to experience anxiety connected with dental procedures. Surveys used in the study conducted prior to the aesthetic procedures enabled identification of patients with a tendency towards perfectionism. Following such a pattern can contribute to better preparation before an aesthetic procedure and may result in greater patient satisfaction and reduced anxiety.

A study was conducted on a group of 406 obese patients (mostly women, aged 29–58 years) to evaluate the association of psychological symptoms with levels of self-esteem before and after stomach Botox treatment [44]. After the procedure, patients achieved mental and physical health comparable to the average health of non-obese patients. Using questionnaires, the relationship between anxiety and respondents’ self-esteem was proven on many levels, which enables healthcare personnel to focus on stress and anxiety reduction in patients before aesthetic medicine procedures. Research concludes that aesthetic medicine procedures have a positive impact, not only on the external appearance of patients, but also on their overall self-esteem after the procedure.

Conclusions

In this study, the comparison of certain artificial intelligence (AI) applications versus self-assessment sheds light on the potential of using new technologies, with the assumption that, for example, the assessment made using those could be more objective, but it may underscore its limitations in the context of cosmetic surgery. The application of AI in facial analysis, while offering objective data, falls short in capturing the nuanced self-perception and psychological dimensions crucial in cosmetic procedures. Nonetheless, further validation with larger and more diverse populations and other AI tools remains necessary to ensure more robust and generalizable findings. Overall, while AI presents promising opportunities for enhancing medical practices, including cosmetic surgery, its integration should be approached with a nuanced understanding of its capabilities and limitations, especially concerning the intricate interplay between physical appearance, psychological factors, and patient expectations.

In addition, integrating AI-based tools with validated psychological screening protocols, such as BDD screening, could refine pre-surgical patient assessments by identifying individuals with heightened risk or unrealistic expectations. A structured, longitudinal framework is also recommended to measure how AI-based simulations influence patient satisfaction across postoperative recovery, thus guiding best practices in surgical planning and patient counseling.

Although artificial intelligence delivers measurable, valuable, and objective data for physical appearance assessment, it does not fully capture the subjective elements of self-perception and the psychological factors involved in aesthetic procedures. The outcomes indicate that relying only on the data provided by AI-based tools concerning facial assessment may not be sufficient for clear discussion of patient’s expectations and self-perception. Therefore, it is essential to involve a panel of professionals who can provide a comprehensive evaluation that integrates both objective AI-driven insights and subjective human perspectives. This multidisciplinary approach ensures that patients receive well-rounded guidance that considers not only measurable aesthetic parameters but also their psychological well-being, personal preferences, and emotional responses to aesthetic changes. By combining technology with expert judgment, practitioners can foster a more meaningful dialog with patients, leading to more informed and satisfactory decisions in aesthetic procedures.

The AI assessment, based on learning algorithms that evaluate facial features on the 7 dimensions produces worse results than the patient’s self-assessment, so it may not represent the internal perceptual structure of the patient. Considering these pilot findings, AI-based pre-treatment facial analysis should not be the exclusive method used to schedule procedures in cosmetic surgery.