Multi-Modal AI Plastic Surgery Diagnosis and Decision-Making System: Integration of Deep Learning and Psychological Assessment

Multi-Modal AI Plastic Surgery Diagnosis and Decision-Making System: Integration of Deep Learning and Psychological Assessment | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Multi-Modal AI Plastic Surgery Diagnosis and Decision-Making System: Integration of Deep Learning and Psychological Assessment Xiaohui Qiu, Shuyue Chen, Jiangjie Tang, Qiuyang Chen, Shenghui Liao, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6233674/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Purpose : To design and develop an AI-based plastic surgery recommendation system using 3D photographs and psychological questionnaire surveys, aiming to provide personalized treatment solutions for the plastic surgery industry. Methods : Based on artificial intelligence technology, this study utilized patients' 3D photographs and psychological questionnaire results as training samples to construct a personalized AI-based plastic surgery recommendation system. This system comprehensively considers factors such as patients' anxiety levels, economic status, and psychological expectations. The study selected 5,543 cases of plastic surgery outpatients aged 18 to 55 years, collected their 3D photographs and questionnaire data, and used these for AI system training. The software predicted treatment projects and compared them with doctors' predictions to validate the system's accuracy and patient satisfaction. Third-party doctors evaluated the system's safety, ultimately developing an efficient and accurate plastic surgery recommendation system. Results : The economic downturn in the post-COVID-19 era significantly impacted psychological health and the plastic surgery industry. Factors such as age, education level, income, and gender had significant effects on patients' psychological state and treatment willingness. The AI system integrated patients' psychological state, gender, income, and physical characteristics, providing personalized plastic surgery treatment suggestions and achieving a 93.25% patient satisfaction rate. Conclusion : The AI-based personalized plastic surgery recommendation system offers an innovative solution for the industry, enhancing the accuracy of treatment suggestions and patient satisfaction, thereby promoting sustainable development. In the post-pandemic era, the plastic surgery industry should focus on patients' physical, psychological, and economic factors to achieve personalized services. Biological sciences/Biological techniques/Imaging Biological sciences/Biological techniques/Software Artificial intelligence COVID-19 epidemic plastic surgery education income gender mental state Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Following the adjustment of China's COVID-19 control policies, the macroeconomy entered a cyclical downturn phase (Q1 2023)(1, 2). Contact-intensive service industries (retail, dining, tourism) experienced operational disruptions and demand-side contractions, leading to a year-on-year decrease of 18.7% in industrial added value and resulting in a typical Keynesian effective demand deficiency phenomenon(3-5). This economic contraction was amplified through the multiplier effect of the "income-consumption-expectation" cycle, with the World Health Organization (WHO) standardizing mental health questionnaire (EQ-5D-5L) showing a 32.4% decline in urban residents' mental health index compared to pre-pandemic levels, while the anxiety disorder prevalence rate rose to 14.8% (6-9). The impact of socio-economic issues on population mental health exhibits significant demographic heterogeneity: the youth group (18-24 years old) faces structural employment contradictions, with the 2023 college graduate survey indicating a 23% decrease in job supply index (PSI) compared to 2019 baseline values, while the job density index (JDI) increased by 17%, forming an expanding asymmetrical distribution of supply-demand gaps (10-12). Meanwhile, the middle-aged group (35-50 years old) faces rigid career transition constraints, with National Bureau of Statistics tracking data showing that the median reemployment cycle for this group extended to 9.8 months (Δ=+127%), significantly higher than the OECD average (Δ=+62%) (OECD, 2023) (13, 14). In this context, the medical aesthetics industry has shown unique development trends. Despite being widely regarded as a sunrise industry, industry data indicates that the market size contracted continuously from 2020 to 2022 (CAGR=-8.9%), and after policy easing in 2023, only a weak recovery of 1.2% was achieved, still failing to return to the 2019 baseline level (15-17). Notably, consumer decision-making mechanisms have undergone systemic changes. Based on a multi-factor regression analysis of 5,300 outpatient cases, our team observed that the acceptance of treatment plans by seekers of aesthetic services is influenced by a combination of factors including gender, age, psychological state, and price sensitivity, with some cases exhibiting cognitive biases in understanding indications (18-20). In clinical practice, both physicians and patients face dual challenges in the decision-making process: on one hand, patients may not fully express their true needs, and on the other hand, the information physicians obtain during limited outpatient visits is often one-sided, making it difficult to comprehensively analyze patients' multi-dimensional needs. Therefore, developing tools that can analyze multi-source information and provide reasonable and safe treatment suggestions has become an important research direction to address these challenges. In response to this complex decision-making scenario, we decided to developed an intelligent assistance system based on multi-modal data fusion. The system integrates four-dimensional data sources: ① high-precision 3D facial point cloud reconstruction model data (RMS=0.1mm); ② standardized psychological assessment tools (EQ-5D-5L/PHQ-9/GAD-7); ③ dynamic economic affordability models; ④ social media interest and attention analysis. Through training using a self-supervised convolutional neural network framework (CAE-SCNN), our team's previous research demonstrates that this algorithmic framework can complete relevant tasks with significantly reduced training samples while maintaining an 83.2% clinical recommendation compliance rate (21, 22), showing a clear advantage over traditional deep learning methods (P<0.01) A bibliometric analysis (Web of Science Core Collection 2020-2024) reveals a knowledge gap in this research area: among 2,438 relevant articles, only 31 related papers were found that involve multi-factor interaction analysis concerning AI, psychology, treatment recommendations, and plastic surgery, while only 6 studies integrate artificial intelligence with clinical decision-making. Furthermore, the field of multi-modal AI decision-making based on psychological, economic, and biological features remains a blank area in plastic surgery (the results of the bibliometric analysis are shown in Figure 1). This study constructs, for the first time, a decision support system that spans psychology, plastic surgery, and software engineering, providing data-driven transformation and upgrading solutions for the industry. The innovation of this research lies in integrating multi-modal data (3D facial point clouds, psychological assessment questionnaires, economic affordability models, and social media profiles) into an aesthetic treatment decision support system, addressing the gap in multi-modal AI decision-making based on psychological, economic, and biological features in plastic surgery. Additionally, through the innovative application of the self-supervised convolutional neural network framework (CAE-SCNN), The training results can still be good based on a smaller sample size., demonstrating a substantial improvement over traditional deep learning methods (P<0.01). This research provides new technical pathways for precision medicine in plastic surgery by combining artificial intelligence algorithms with psychological, economic, and biological features. In terms of plastic surgery development, this study has significant theoretical and practical implications. Theoretically, it expands the research boundaries of decision support systems in plastic surgery and offers a new research paradigm for the application of multi-modal data fusion and AI algorithms in medicine. Practically, the intelligent assistance system developed in this study enables clinicians to more comprehensively understand patient needs, optimize personalized treatment plans, and improve the efficiency and quality of medical services. Furthermore, by providing data-driven solutions for the digital transformation of the medical aesthetics industry, this research supports the industry's transition from traditional models to intelligent and precise approaches, injecting new vitality into the sustainable development of plastic surgery. Methods 1. Ethics & Privacy Protection This study strictly adhered to the principles outlined in the Declaration of Helsinki, China's Ethical Review Measures for Biomedical Research Involving Humans, and the European Union's General Data Protection Regulation (GDPR) to ensure ethical compliance throughout the research process and to maximize the protection of participants' rights. Specific measures are detailed below: 1.1 Ethics approval: Approved by the Ethics Committee of Xiangya Third Hospital (No. K23688) with independent monitoring; 1.2 Informed consent: Three-level consent forms detailed data usage, anonymization, and withdrawal rights; the facial photos used in this study and the collected questionnaire data have been obtained with informed consent from all participants, and the participant in Figure 5 has agreed to publish in an open access journal. 1.3 Privacy measures :Differential privacy (ε=0.76, Laplace noise Δf=1.2) with k-anonymity (k=5) and l-diversity (l=2); AES-256 encrypted storage on private cloud with two-factor authentication; "Safe harbor" datasets containing only aggregated statistics were shared externally; 1.4 Risk control : Independent Data Monitoring Committee (IDMC) conducted quarterly audits to prevent unauthorized access. 2. Multimodal Database Construction A multicenter cross-sectional study was conducted from December 1, 2023, to December 1, 2024, across 33 plastic surgery clinics in Changsha. Participants (n=5,543, response rate: 93.72%) were enrolled with the following data collected: 2.1 Psychological assessment : Standardized scales (EQ-5D-5L, PHQ-9, GAD-7, ISI, IES-R, HRQOL) and novel scales (COVID-19-related Plastic Surgery Intention Scale, Trauma Acceptance Scale); 2.2 3D facial imaging : Captured using iPhone 14 Pro under standardized conditions (object distance: 1.2m, aperture: f/2.8, depth-enhanced mode, dual-lens depth sensing); 2.3 Clinical data : Physician recommendations, patient decisions, and postoperative outcomes (satisfaction, recovery time, re-visit rate) were automatically recorded via an electronic medical record (EMR) system. 3. Data Preprocessing and Modeling 3.1 Feature Engineering 3.1.1 Ordinal variables (e.g., insomnia severity): Encoded with ordinal labeling; 3.1.2 Nominal variables (e.g., gender, occupation): Processed via one-hot encoding; 3.1.4 Continuous variables (e.g., age, income): Normalized using Z-score; 3.2.4 Text data: Tokenized via Jieba and vectorized with TF-IDF for BiLSTM input. 3.2 Self-supervised Multimodal Fusion A two-phase training framework was implemented to reduce annotation dependency: 3.2.1 Self-supervised pretraining : ① Contrastive learning (SimCLR variant) applied to unlabeled 3D point clouds and text data; ② Augmentations: Random rotation (±15°), occlusion (20% regions), and Gaussian noise (σ=0.01) for point clouds; synonym replacement (WordNet) for text; ③ Loss function: NT-Xent loss with negative pair margin set to 1.2. 3.2.2 Multimodal fusion : ① Architecture: ResNet-50 for geometric features, BiLSTM for text, and FC-512 for clinical data; ② Cross-modal attention: 4-head attention dynamically weighted modality contributions; ③ Transfer learning: Frozen pretrained encoders reduced supervised data requirements by 63%. 3.2.3 Class Balancing SMOTE-ENN hybrid sampling improved minority class (e.g., severe depression, n=227) F1-score from 0.68 to 0.81 (Δ=+19.1%) while reducing overfitting risk by 22%. 3. 3D Point Cloud Reconstruction & QC ① Point cloud generation: Agisoft Metashape (v2.0) produced dense clouds (mean reprojection error: 0.23mm, SD=0.07); ② Illumination correction: Spherical harmonics decomposed diffuse/specular components; ③ Quality control: CloudCompare evaluated density (>500 pts/cm²) and curvature continuity (Hausdorff distance <0.5mm), excluding outliers (n=43). 4. Model Training & Validation 4.1 5-fold cross-validation: 5,543 cases were stratified (train/validation=8:2); 4.2 Loss function: Class-weighted cross-entropy (weights=1/√N); 4.3 Optimization: AdamW (lr=3e-5, β1=0.9, β2=0.999) with cosine annealing scheduler (T_max=50). 5. Clinical Double-blind Validation 5.1 Evaluation cohort : 150 new patients independently assessed by 3 chief physicians (mean experience: 18.6 years) and the AI system; 5.2 Metrics : Patient satisfaction: Likert 5-point scale (Cronbach's α=0.89); Physician ratings: Treatment accuracy (ICC=0.92) and safety (NCCN compliance ≥95%); Statistical Analysis Data analysis was performed using SAS University Edition (v3.8), with a statistical significance threshold set at two-tailed α = 0.05. All tests excluded multicollinearity interference through variance inflation factors (VIF < 5). The specific methods are as follows: 1. Between-group Comparative Analysis 1.1 Accuracy Comparison : For the accuracy scores of treatment plans provided by physicians and the AI system (continuous variables), paired-sample t-tests were used to compare group differences, with the effect size quantified using Cohen's d values (normality confirmed by Shapiro-Wilk test, p > 0.1). 1.2 Safety Assessment : For NCCN guideline compliance rates (binary variables), McNemar tests (paired chi-square tests) were applied to calculate odds ratios (OR) and 95% confidence intervals. 1.3 Patient Satisfaction : Based on a Likert 5-point scale (ordered categorical variable), Wilcoxon signed-rank tests were used to analyze group differences, and Hodges-Lehmann median offsets were reported. 2. Non-inferiority Test Design A non-inferiority margin (Δ = 10%) was predefined. For the primary efficacy metrics (e.g., treatment accuracy) between the AI system and physicians, two-sided Z-tests were used to calculate 95% confidence intervals for group differences. Non-inferiority was established if the upper bound of the confidence interval was below Δ. 3. Correlation and Predictive Modeling 3.1 Categorical Variable Associations: The relationships between gender (binary), education level (ordered categorical), and post-COVID-19 psychological metrics (continuous PHQ-9/GAD-7/EQ-5D-5L/PHQ-9/GAD-7/ISI/IES-R/HRQOL scores) were analyzed using: 3.2 Spearman rank correlation analysis (education level vs. psychological metrics). Mann-Whitney U tests (gender group comparisons). 3.3 Plastic Surgery Intention Prediction: A multivariate logistic regression model was constructed, with plastic surgery intention as the outcome variable. Covariates included age, gender, and anxiety/depression scores. Variables were selected using stepwise regression (AIC criterion), and model performance was evaluated using the area under the ROC curve (AUC). 4. Statistical Control and Correction 4.1 Multiple Comparisons: Bonferroni corrections were applied to address multiple testing issues, adjusting the significance threshold to α' = 0.05/k (k = number of tests). 4.2 Missing Data Handling: Missing values for continuous variables (<5%) were addressed using multiple imputation (PROC MI), while missing values for categorical variables were replaced by the mode. 5. Sensitivity Analysis 5.1 Robustness Verification: Bootstrap resampling (1,000 iterations) was used to validate the robustness of primary findings. 5.2 Non-normal Data: For non-normally distributed data (Kolmogorov-Smirnov test p < 0.05), robust regression (Huber-White standard errors) was applied in secondary analyses. Results The overall process flow diagram can be seen in Figure 2. 1. Text data analysis (including questionnaires, diagnostic recommendations, and demographic baseline) and quality assessment of the text dataset. The statistical data on the participants' basic information is shown in Table 1; the statistical data on the participants' psychological state scores is shown in Table 2. Table 1：participants' basic information Age Gender Education Level Income Grouping Youth Group （18-30） Middle-Aged Group（30-60） Male Female Highly Educated Group Low Education Group Highly income gourp Low income group Number of People 3310 1885 879 4316 3810 1385 1681 3514 Percentage 63.72% 36.28% 16.92% 83.08% 73.34% 26.66% 32.36% 67.64% Talbe2：participants' psychological state scores Population grouping EQ-5D-5L PHQ-9 GAD-7 ISI IES-R HRQOL Intention for plastic surgery Youth Group（mine±SD) 0.772±0.825 0.456±1.791 1.112±1.811 1.238±1.690 17.042±9.484 5.899±4.253 7.765±0.477 Middle-Aged Group（mine±SD) 0.656±0.918 0.164±0.761 1.337±2.602 1.873±2.424 24.373±5.07 10.019±4.601 6.629±1.184 P value 0.000 0.000 0.000 0.000 0.000 0.033 0.000 male（mine±SD) 0.173±0.695 0.554±1.878 0.272±0.699 0.126±0.466 6.887±6.572 3.386±3.997 7.958±0.265 female（mine±SD) 0.842±0.849 0.307±1.409 1.383±2.282 1.747±2.099 22.374±6.735 8.247±4.545 7.220±1.022 P value 0.000 0.000 0.000 0.000 0.006 0.000 0.000 Highly education lever（mine±SD) 0.743±0.849 0.379±1.577 1.128±1.997 1.372±1.873 19.397±9.041 7.381±4.872 7.422±0.922 Low education lever（mine±SD) 0.635±0.945 0.136±0.795 1.646±2.896 2.154±2.707 22.191±7.164 7.719±4.398 6.819±1.169 P value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Highly income group（mine±SD) 0.787±1.014 0.164±0.824 1.453±2.646 1.901±2.402 23.444±6.320 9.792±4.494 6.776±1.167 Low-income group（mine±SD) 0.684±0.718 0.494±1.859 0.992±1.605 1.135±1.568 16.837±9.491 5.549±4.195 7.795±0.421 P value <0.001 <0.001 <0.001 <0.001 <0.001 0.489 0.000 The analysis of questionnaire data mining revealed that differences in anxiety, depression, and intention for cosmetic surgery after COVID-19 among different age groups (see Table 2 part 1). Specifically, the EQ-5D-5L health scale indicates that compared to the middle-aged group, the younger group has a poorer self-assessed health status; the results from the PHQ-9 depression screening scale show that the younger group has a higher severity of self-assessed depression compared to the middle-aged group; the Generalized Anxiety Disorder 7-item scale (GAD-7) indicates that the middle-aged group has higher anxiety levels than the younger group; the Insomnia Severity Index (ISI) shows that the middle-aged group experiences higher levels of insomnia than the younger group; the Impact of Event Scale-Revised (IES-R) demonstrates that the middle-aged group is significantly more affected by the epidemic event than the younger group; the Health-Related Quality of Life scale (HRQOL) reflects that due to COVID-19, the middle-aged group's self-assessed health-related quality of life is poorer than that of the younger group; the intention for cosmetic surgery scale shows that the younger group has a higher negative intention towards cosmetic surgery than the middle-aged group, indicating that the middle-aged group has a higher intention for cosmetic surgery. All these differences have p-values less than 0.05, indicating statistical significance. Additionally, the study also found relationships between anxiety, depression, and intention for cosmetic surgery after COVID-19 with variables such as gender, education level, and income (see Table 2 part2，3，4). Specific details include differences in health status, depression severity, anxiety levels, insomnia severity, epidemic impact, and intention for cosmetic surgery among different gender, education level, and income groups, with most differences being statistically significant. It is worth noting that the p-value for the Health-Related Quality of Life scale (HRQOL) is greater than 0.05, indicating that the difference in income is not statistically significant, while differences among other variables are statistically significant. We further analyzed the relationships between psychological factors, gender, age, education, income, and their impact on the intention for plastic surgery decision-making in the post-pandemic era, as shown in Figure 3. The correlation trends between anxiety, depression, and intention for plastic surgery with different age groups are shown in (Figure 3, A). The correlation coefficients between age and various factors at a 95% confidence interval are as follows: For EQ-5D-5L, R = -0.05033, with a p-value of 0.0003. For PHQ9, R = -0.08541, with a p-value less than 0.0001. For GAD-7, R = 0.05986, with a p-value less than 0.0001. For ISI, R = 0.1563, with a p-value less than 0.0001. For IES-R, R = 0.3816, with a p-value less than 0.0001. For HRQol, R = 0.3182, with a p-value less than 0.0001. For Intention for plastic surgery, R = -0.5295, with a p-value less than 0.0001. Anxiety, depression, and intention for plastic surgery trends with different income levels are shown in (Figure 3, B). The correlation coefficients between income and various factors at a 95% confidence interval are as follows: For EQ-5D-5L, R = 0.03392, with a p-value of 0.0145. For PHQ9, R = -0.09332, with a p-value less than 0.0001. For GAD-7, R = 0.1, with a p-value less than 0.0001. For ISI, R = 0.1892, with a p-value less than 0.0001. For IES-R, R = 0.4318, with a p-value less than 0.0001. For HRQol, R = 0.4302, with a p-value less than 0.0001. For Intention for plastic surgery, R = -0.677 with a p-value less than 0.0001. The correlation trends between anxiety, depression, and intention for plastic surgery with different genders are shown in (Figure 3, C). The correlation coefficients between genders and various factors at a 95% confidence interval are as follows: For EQ-5D-5L, R = 0.2907, with a p-value less than 0.0001. For PHQ9, R = -0.06165, with a p-value less than 0.0001. For GAD-7, R = 0.1945, with a p-value less than 0.0001. For ISI, R = 0.3012, with a p-value less than 0.0001. For IES-R, R = 0.6544, with a p-value less than 0.0001. For HRQol, R = 0.3783, with a p-value less than 0.0001. For Intention for plastic surgery, R = -0.2829 with a p-value less than 0.0001. The correlation trends between anxiety, depression, and intention for plastic surgery with different educational backgrounds are shown in (Figure 3, D). The correlation coefficients between education lever and various factors at a 95% confidence interval are as follows: For EQ-5D-5L, R = 0.0197, with a p-value =0.1558. For PHQ9, R = 0.02311, with a p-value =0.0959. For GAD-7, R = -0.00967, with a p-value=0.4860. For ISI, R = -0.0276, with a p-value=0.0744. For IES-R, R = 0.0344, with a p-value=0.0132. For HRQol, R = 0.1467, with a p-value less than 0.0001. For Intention for plastic surgery, R = -0.0519 with a p-value less than 0.0001. The correlation trends between anxiety levels and intention for plastic surgery, as well as depression levels, are shown in (Figure 3, E). The correlation coefficients between anxiety levels and various factors at a 95% confidence interval are as follows: For EQ-5D-5L, R = 0.0275, with a p-value =0.1349. For PHQ9, R = 0.03992, with a p-value =0.004. For ISI, R = 0.7265, with a p-value less than 0.0001. For IES-R, R = 0.1312, with a p-value less than 0.0001.. For HRQol, R = 0.03366, with a p-value= 0.0153. For Intention for plastic surgery, R = -0.1608 with a p-value less than 0.0001. Based on the survey data analysis in the preceding text, we have clarified the significant correlation between psychological status and demographic characteristics in decision-making for plastic surgery. To further investigate this, we constructed a logistic regression model incorporating 12 predictive factors, systematically integrating the following dimensions: Core psychological scales : Including depression (PHQ9), anxiety (GAD-7), insomnia (ISI), and post-traumatic stress (IES-R). Demographic characteristics : Covering age, gender, income, and education level. Health status indicators : Including the health-related quality of life scale (HRQol) and the EuroQol 5-dimensional scale (EQ-5D-5L). Based on the analysis of the correlations in the aforementioned questionnaire data, we can clearly determine that factors such as age, gender, education level, income, and various psychological dimensions significantly influence patients' intentions and decision-making regarding plastic surgery. Therefore, it is essential to incorporate data from patient psychological questionnaires and baseline surveys into the deep learning dataset when developing an AI-based plastic surgery recommendation system in order to construct a multidimensional predictive model. During the model performance evaluation phase, we validated the predictive capabilities of the training and testing sets through ROC curve analysis. As shown in Figure 4, the AUC value for the training set reached 0.96, while the AUC value for the testing set was 0.95. The high similarity between these two curves indicates that the model demonstrates good stability and reliability. This result further confirms the superior performance of the multidimensional model in predicting patients' intentions regarding plastic surgery treatment decisions. 2. 3D point cloud generation from facial photo data Facial information is the cornerstone of surgical treatment and consultation decisions. Therefore, we adopted standardized photography techniques to meticulously capture patient facial features and converted the 3D photographs into 3D point cloud data, providing a richer source of information for deep learning software. The standard photographs and point cloud models captured in our study are shown in Figure 5. In the outpatient setting, the standardized photography of patient facial images is shown in Figures 5 A, B, and C. We used Agisoft Metashape software to perform 3D scanning of the captured images, generating an STL 3D model, which was then subjected to topological optimization to ensure that the number of triangular facets remained below 50,000, followed by the creation of a dense point cloud. By downsampling the point cloud, we converted it into a standard point cloud model for deep learning in this study, and further validated the point cloud file using CloudCompare software. The specific steps included: ICP registration comparison (registration error ≤ 0.05 mm), outlier removal (threshold set to 3σ standard deviation), and normal consistency optimization (angle threshold of 15°). This process retained all 21 key facial landmark points, including the tip of the nose and inner canthus points. Ultimately, the point cloud model achieved a density of 40% (equivalent to a resolution of 0.2 mm), with a total of 19,300 effective data points (distribution density variance < 8%), and the model file size was compressed to 35% of the original data. Additionally, non-soft tissue interference points, such as hair (accounting for < 5%), were also removed (see Figure 5D). 3. Comparison of the AI Plastic Surgery Diagnostic and Treatment Recommendation System with Physician Diagnostic Recommendations After completing the training of the AI plastic surgery diagnostic and recommendation system, we re-enrolled 150 outpatient patients, collecting patient questionnaires and facial information. We then conducted both AI-based plastic surgery diagnostics and recommendations and manual diagnostics and recommendations (provided by three senior chief physicians). Subsequently, we performed a double-blind evaluation of the diagnostic recommendations by a third-party physician, assessing aspects such as the accuracy and safety of the recommendations, as well as patient satisfaction with the diagnostic results. The comparative study and analysis of the results are detailed in Figure 6 and Tables 3, 4, and 5. Table 3. Accuracy of treatment recommendations project AI group (n=150) doctor group (n=150) Non inferiority test results“Δ=0.2” ‌Mean (SD) ‌ 4.25 (0.16) ‌ 4.18 (0.22) ‌ — ‌ Paired t-test P-value 0.054 ‌ — ‌ Cohen's d effect quantity 0.19 ‌ — ‌ Non inferiority Z-value & P-value — — Z=2.12 & P=0.017 ‌ The non-inferiority threshold Δ is set at 0.2 points (based on clinical equivalence consensus) 23 Conclusion: The accuracy of the AI group is not inferior to that of the physician group (P=0.017), and there is no statistically significant difference in the mean between the two groups (P=0.054) Table 4. Comparison of Safety Project AI group (n=150) doctor group (n=150) Non inferiority test results“Δ=5%” ‌ Safety rate of diagnosis and treatment recommendations 86.0% ‌ 88.7% ‌ — ‌ McNemar test P-value 0.124 ‌ — ‌ Odds ratio (OR) 0.81 (95% CI: 0.58–1.13) ‌ — ‌ Non inferiority Z-value/P-value — — Z=1.76/P=0.039 ‌23 The non-inferiority threshold Δ is set at 5% (based on clinical safety consensus) Therefore the safety of the AI group was not inferior to that of the physician group (P=0.039), and the difference between the two groups was not statistically significant (P=0.124) Table 5. Patient satisfaction of treatment suggestion evaluation Project AI group (n=150) doctor group (n=150) ‌Mean (SD)‌ 4.21 (0.15) ‌ 3.79 (0.25) ‌ ‌Wilcoxon P value‌ <0.001 ‌ ‌ Hodges Lehmann offset ‌ 0.42 ‌ — In terms of the accuracy of treatment recommendations, the AI group had an average score of 4.25 (SD=0.16), slightly higher than the doctor group’s score of 4.18 (SD=0.22). The paired t-test results showed a P-value of 0.054, indicating that the difference between the two groups is not statistically significant. Furthermore, the non-inferiority test (Z=2.12, P=0.017, non-inferiority threshold Δ=0.2) further confirmed that the accuracy of the AI group is not inferior to that of the doctor group. Therefore, we can conclude that the accuracy of the AI group is clinically equivalent to that of the doctor group, with no significant difference in the means between the two groups. Regarding safety, the NCCN guideline compliance rate for the AI group was 86.0%, slightly lower than the 88.7% for the doctor group. The McNemar test yielded a P-value of 0.124, indicating that the difference between the two groups is not statistically significant. The non-inferiority test (Z=1.76, P=0.039, non-inferiority threshold Δ=5%) showed that the safety of the AI group is not inferior to that of the doctor group. This indicates that the safety of the AI group is comparable to that of the doctor group, with no statistically significant difference between the two groups. In terms of patient satisfaction, the AI group had an average score of 4.21 (SD=0.15), significantly higher than the doctor group’s score of 3.79 (SD=0.25). The Wilcoxon test yielded a P-value of <0.001, indicating that the difference between the two groups is statistically significant. The Hodges-Lehmann offset was 0.42, further confirming that patient satisfaction in the AI group is significantly higher than in the doctor group, with a statistically significant difference. In summary, the AI plastic surgery diagnostic and recommendation system demonstrates excellent performance in terms of accuracy, safety, and patient satisfaction. Its accuracy and safety are comparable to those of the doctor group, while it significantly outperforms the doctor group in patient satisfaction. This indicates that the AI system has high clinical application value in plastic surgery diagnostic recommendations, providing patients with a superior service experience. Discussion This study validates the outstanding performance of an AI-based plastic surgery diagnostic recommendation system that integrates multi-modal data fusion in terms of accuracy, safety, and patient satisfaction. Comparative analysis with the physician group shows that the AI group performs comparably in treatment recommendation accuracy (mean difference not statistically significant, P=0.054) and safety (non-inferiority test passed, P=0.039), while significantly outperforming the physician group in patient satisfaction scores (P<0.001). These results indicate that the AI system holds high clinical application value in plastic surgery diagnostic recommendations, providing a potential for technological innovation in the field. Multiple studies have shown that gender and income levels significantly influence beauty consumption behavior (22-24), suggesting that the development of plastic surgery diagnostic recommendation systems should differ from other intelligent diagnostic systems, particularly by incorporating psychological assessment factors as reference models. This study further reveals the moderating effects of age, education level, and psychological state on plastic surgery intentions and treatment decisions through multi-modal data analysis. Compared to research on appearance anxiety and social media related to plastic surgery decision-making (25-27)，this study constructs a more comprehensive and precise predictive model by integrating 3D facial imaging, psychological assessment data, and economic models. This innovative approach not only enriches the existing literature but also provides a new research perspective for the intelligent development of the plastic surgery field. By utilizing multi-modal data fusion (including 3D facial imaging, psychological assessment scales, and demographic models), this study successfully developed an efficient AI diagnostic recommendation system, pioneering a new technological pathway in plastic surgery. The introduction of the AI system not only helps physicians better understand patient needs but also optimizes personalized treatment plans, enhancing the efficiency and quality of medical services. The findings provide practical guidance for the digital transformation of the plastic surgery industry, specifically in the following areas: 1. Adapting to Changing Customer Perceptions : In the post-pandemic era, emphasizing health and natural beauty through transparent education and social media can enhance patient trust. 2. Addressing Price Sensitivity: Adjusting treatment strategies based on patients' economic income can attract budget-conscious consumers during economic downturns. 3. Promoting Remote Consultation Services : Utilizing AI technology for contactless consultations can enhance patient acceptance and safety through online consultations and simulation technologies. Limitations and Future Directions Despite the significant progress made in this study, several limitations remain: 1.Sample Limitations : The study's sample is primarily concentrated in specific regions, which may limit the generalizability of the results. Future research should expand the sample range to include diverse geographic and cultural backgrounds to validate the robustness of the model. 2.Model Generalization : The current model's training data is based on a specific population; future efforts should focus on optimizing algorithms to adapt to diverse clinical scenarios. 3.Long-term Effect Evaluation: This study only conducted a short-term observation. and has not assessed the sustained impact of the AI system in long-term clinical applications or the dynamic changes in patient satisfaction. To further optimize the research, future efforts could focus on: 1.Expanding Sample Size : Increasing the sample size and geographic coverage to validate the robustness and applicability of the model. 2.Optimizing Algorithm Framework : Exploring more advanced AI algorithms, such as deep learning and transfer learning techniques, to further enhance the model's predictive accuracy and generalization capabilities(28, 29). 3. Longitudinal Follow-up Studies : Conducting follow-up studies to evaluate the long-term effects of the AI system in clinical applications, including patient satisfaction, treatment outcome stability, and the sustained impact on healthcare efficiency (30). Conclusion This study successfully constructed an efficient AI plastic surgery diagnostic recommendation system through multi-modal data fusion and a self-supervised learning framework, providing innovative solutions for the digital transformation of the plastic surgery field. The results indicate that the AI system significantly enhances diagnostic accuracy, optimizes treatment plans, and improves patient experience. Future research will further refine the technical framework and expand its application scope, supporting the sustainable development of the plastic surgery industry while offering patients more efficient and personalized medical services. Declarations Informed Consent Statement The facial photos used in this study and the collected questionnaire data have been obtained with informed consent from all participants, and the participant in Figure 5 has agreed to publish in an open access journal. Conflict of interest statement The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Author Contribution Xiaohui Qiu: Conceptualized the study, designed the research framework, and led the integration of deep learning with psychological assessment in the multi-modal AI system. Supervised the entire project and finalized the manuscript.Shuyue Chen: Conducted psychological assessments, collected and analyzed clinical data, and contributed to the interpretation of psychological evaluation results to ensure the system's applicability in patient care.Jiangjie Tang: Designed and implemented the biological experiments, including wound healing assays and histological analysis, ensuring the accuracy and reliability of biological data.Qiuyang Chen: Developed and optimized the deep learning algorithms, focusing on model training and validation to ensure the system's diagnostic accuracy and decision-making efficiency.Shenghui Liao: Integrated the AI components with psychological assessment tools, developed the user interface, and ensured the system's functionality and user-friendliness.Jianda Zhou: Conceptualized the study, designed the research framework, and led the integration of deep learning with psychological assessment in the multi-modal AI system. Supervised the entire project and finalized the manuscript.All authors reviewed and approved the final version of the manuscript. Acknowledgement The authors greatly acknowledge the financial support from the National Natural Science Foundation of China (No. 52305313, No .52375341, No.81872219), FuRong Laboratory, Changsha 410078, Hunan, China（No.2023SK2115). Data Availability The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions. References Wu J, Zhan X, Xu H, Ma C (2023): The economic impacts of COVID-19 and city lockdown: Early evidence from China. Structural change and economic dynamics . 65:151-165. Shen Y, Sun A, Zhou Z, Jia D (2024): Digital finance and wealth inequality: Evidence from a big tech platform in China during the COVID-19 pandemic. Pacific-Basin Finance Journal . 83:102226. Jean-françois Piluso N (2025): The Critical Mistake of New Keynesian Economics, Demonstrated in a Revised Model of Blanchard and Kiyotaki. Review of Political Economy .1-16. Kiviruusu O, Ranta K, Lindgren M, Haravuori H, Silén Y, Therman S, et al. (2024): Mental health after the COVID-19 pandemic among Finnish youth: a repeated, cross-sectional, population-based study. The Lancet Psychiatry . 11:451-460. Bistagnino F, Subramanian A, Tovani-Palone MR (2025): Navigating the (Post-) Pandemic Landscape: An Analysis of COVID-19’s Current Status and Future Implications. Disaster Medicine and Public Health Preparedness . 19:e44. Jutakeo J, Kuptniratsaikul V, Limampai P, Akewanlop C, Hantaweepant C, Thephamongkhol K, et al. A Preliminary Report of the Effects of Dharma Creative Art Therapy on Psychological Impacts and Quality of Life of Thai Cancer Patients: A Non-randomized Trial. Machrumnizar M, Muyana N, Bachtiar A, Kusumaratna RK, Suyanto J (2023): Impact of co-epidemic tuberculosis and diabetes mellitus on health-related quality of life: A review. Jurnal Aisyah: Jurnal Ilmu Kesehatan . 8. Wang G, Sabran K (2024): Assessing depression and anxiety among young adults after epidemics and pandemics: a cross-sectional study in Anyang, China. Scientific Reports . 14:2759. Hall BJ, Li G, Chen W, Shelley D, Tang W (2023): Prevalence of depression, anxiety, and suicidal ideation during the Shanghai 2022 Lockdown: A cross-sectional study. Journal of affective disorders . 330:283-290. Mei DW (2023): New changes and challenges in the youth employment in China after COVID-19. Asian Soc Sci . 19:84. Yu Z, Liu L, Zhang X (2024): Bridging the gap: Enhancing employment opportunities for Normal University graduates in China’s knowledge economy. Journal of the Knowledge Economy .1-38. Zheng S, Yan Y (2024): Changes in employment psychology of Chinese university students during the two stages of COVID-19 control and their impacts on their employment intentions. Frontiers in Psychology . 15:1447103. Ayhan F, Elal O (2023): The IMPACTS of technological change on employment: Evidence from OECD countries with panel data analysis. Technological Forecasting and Social Change . 190:122439. Cruz MD (2023): Labor productivity, real wages, and employment in OECD economies. Structural Change and Economic Dynamics . 66:367-382. Li Z, Li T, Zhang L, Zhu Y, Yu Z, Song B (2024): Impact of Public Health and Social Measures on Cosmetic Treatments in the COVID‐19 Pandemic: A Retrospective Multi‐Center Study Combined With a Questionnaire‐Based Cross‐Sectional Study. Journal of Cosmetic Dermatology . 23:3800-3808. Yoon S, Kim YA (2024): Aesthetic Plastic Surgery Issues During the COVID-19 Period Using Topic Modeling. International Journal on Advanced Science, Engineering & Information Technology . 14. Qiao Z, Deng Y, Wang X, Sun Y, Xiong X, Meng X, et al. (2023): The impact of COVID-19 on plastic and reconstructive surgery in China: A single-centre retrospective study. Journal of Plastic, Reconstructive & Aesthetic Surgery . 76:160-168. Zhang Y, Jiang M, Liu J, Liu B (2024): Pursuing beauty: socio-cultural and labor-economic determinants of cosmetic surgery consideration among female college students in China. BMC psychology . 12:519. Mohammed DI, Ibrahim RH (2023): Exploring the impact of psychological factors on cosmetic surgery acceptance: A cross-sectional study. Informatics in Medicine Unlocked . 39:101231. Rahman E, Rao P, Webb WR, Garcia PE, Ioannidis S, Tam E, et al. (2024): Integrating Psychological Insights into Aesthetic Medicine: A Cross-Generational Analysis of Patient Archetypes (IMPACT Study). Aesthetic Plastic Surgery .1-15. Qiu X, Zhong C, Chen Q, Zhao Y, Yang T, Zhou J, et al. (2025): Zygomatic Osteotomy surgery design software based on skull CT scans-Self-supervised algo reduces workload. Journal of Cranio-Maxillofacial Surgery . Qiu X, Han W, Dai L, Zhang Y, Zhang J, Chai G, et al. (2022): Assessment of an Artificial Intelligence Mandibular Osteotomy Design System: A Retrospective Study. Aesthetic Plastic Surgery . 46:1303-1313. Li Z, Cao H, Yu N, Qin F, Li Y, Li Z, et al. (2023): What factors will influence patients when choosing plastic surgeons? A behaviour analysis of Chinese patients. Journal of Plastic, Reconstructive & Aesthetic Surgery . 83:57-68. Chou DW, Layfield E, Prasad K, Shih C, Brandstetter K (2023): Gender and ethnic diversity in academic facial plastic surgery. The Laryngoscope . 133:1869-1874. Mironica A, Popescu CA, George D, Tegzeșiu AM, Gherman CD (2024): Social media influence on body image and cosmetic surgery considerations: a systematic review. Cureus . 16. KÜÇÜK ÖZTÜRK G, BAŞER E, ÖZOCAK H (2023): Investigation of Basic Personality Traits and Social Appearance Anxiety of Individuals Undergoing Aesthetic Surgery Operations: A Descriptive Study. Turkiye Klinikleri Journal of Nursing Sciences . 15. Tokgöz E, Carro MA (2023): Engineering psychology of facial plastic surgery patients. Cosmetic and reconstructive facial plastic surgery: A review of medical and biomedical engineering and science concepts : Springer, pp 343-365. Chen J, Huang R, Chen Z, Mao W, Li W (2023): Transfer learning algorithms for bearing remaining useful life prediction: A comprehensive review from an industrial application perspective. Mechanical Systems and Signal Processing . 193:110239. Hakim M, Omran AAB, Ahmed AN, Al-Waily M, Abdellatif A (2023): A systematic review of rolling bearing fault diagnoses based on deep learning and transfer learning: Taxonomy, overview, application, open challenges, weaknesses and recommendations. Ain Shams Engineering Journal . 14:101945. Yan P, Guo J, Su X, Bai C (2024): Long-Term Tracking of Evasive Urban Target Based on Intention Inference and Deep Reinforcement Learning. IEEE Trans Neural Netw Learn Syst . 35:16886-16900. Additional Declarations No competing interests reported. Supplementary Files completequestionnaire.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6233674","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":436274409,"identity":"75a3808a-ed5e-403d-b119-a8bff5f7cf12","order_by":0,"name":"Xiaohui Qiu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyUlEQVRIiWNgGAWjYBAC+4bjBw5IVNjwMLY3EKnFgPFM4gOLM2kyzD0HiNXCfMDYoLLlsA37jAQitZizHUiTuNlwmId35uONNxhqbKIJarHsOXhMcuaOdB7J2WnFFgzH0nIbCOq5cSBNWvKMNY/h7BwzCcaGw0Rouf/ATPpvGzOP/c0zRGoxOAD0vmSbMw/jDB4itUg2AANZ4kwaD2MP0C8JxPiFnwESlfaM7Yc33vhQY0OEX5AdKZFAinKIFlJ1jIJRMApGwcgAANlmRQ1zlQQTAAAAAElFTkSuQmCC","orcid":"","institution":"Third Xiangya Hospital","correspondingAuthor":true,"prefix":"","firstName":"Xiaohui","middleName":"","lastName":"Qiu","suffix":""},{"id":436274410,"identity":"a117c5be-c65a-47f7-bd24-67692eb2eda4","order_by":1,"name":"Shuyue Chen","email":"","orcid":"","institution":"Third Xiangya Hospital","correspondingAuthor":false,"prefix":"","firstName":"Shuyue","middleName":"","lastName":"Chen","suffix":""},{"id":436274411,"identity":"4f766b37-a251-402d-a196-aa42d3ae6102","order_by":2,"name":"Jiangjie Tang","email":"","orcid":"","institution":"Third Xiangya Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jiangjie","middleName":"","lastName":"Tang","suffix":""},{"id":436274412,"identity":"53e4f640-47c4-478c-976b-671b70281406","order_by":3,"name":"Qiuyang Chen","email":"","orcid":"","institution":"Central South University","correspondingAuthor":false,"prefix":"","firstName":"Qiuyang","middleName":"","lastName":"Chen","suffix":""},{"id":436274413,"identity":"fdf09c16-0b7d-4add-8e51-e0ceb01e2cee","order_by":4,"name":"Shenghui Liao","email":"","orcid":"","institution":"Central South University","correspondingAuthor":false,"prefix":"","firstName":"Shenghui","middleName":"","lastName":"Liao","suffix":""},{"id":436274414,"identity":"cf324936-5968-4c2a-8ccb-70b0077b6ec0","order_by":5,"name":"Jianda Zhou","email":"","orcid":"","institution":"Third Xiangya Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jianda","middleName":"","lastName":"Zhou","suffix":""}],"badges":[],"createdAt":"2025-03-15 15:23:07","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6233674/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6233674/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":80089534,"identity":"7a9a536c-f6df-4e18-a3e7-5f6ed20a0abd","added_by":"auto","created_at":"2025-04-07 18:16:46","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":320187,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBibliometric analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003eFig. 1: The figure illustrates a network graph constructed using four keywords: social group mental health, artificial intelligence, treatment recommendation, and plastic surgery. The connections between nodes reveal the intricate relationships among these keywords. The size of each node potentially represents the frequency or importance of the keyword within the literature dataset, while the thickness of the lines may reflect the strength of the association between the keywords. As the figure indicates, selecting any three of the four keywords yields a relatively large number of related studies. However, only six studies incorporate all four keywords, and their primary focus differs significantly from that of the present study. Therefore, the novelty of this research lies in the establishment of a large-sample, multi-modal fusion database encompassing psychological, facial feature, and demographic data, upon which a deep learning study based on convolutional neural networks was conducted.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6233674/v1/0c0a7eaec80c3a7e59e3d00f.png"},{"id":80089839,"identity":"4d4aeae8-5e5b-4567-946c-daf090ebf567","added_by":"auto","created_at":"2025-04-07 18:24:46","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":159271,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eOverall Design \u0026amp; Algorithm Logic Diagram\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFig 2.\u003c/strong\u003e The entire study involved collecting psychological questionnaires, baseline data, 3D facial photographs, and diagnostic treatment outcomes from a large number of outpatients. After data processing and feature extraction, a multi-source fusion database was established. This database served as a deep learning sample library and was applied to the team's self-developed deep learning algorithm framework. After training, the accuracy, safety, and patient satisfaction of the diagnostic and treatment recommendations output by the algorithm were evaluated.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6233674/v1/0c9cf7d3fa8a039a8a6a5e74.png"},{"id":80089929,"identity":"ce04bf41-d8df-4553-a90b-efbb3358ed16","added_by":"auto","created_at":"2025-04-07 18:32:46","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":208344,"visible":true,"origin":"","legend":"\u003cp\u003eThe correlation between psychological questionnaire results and the intention for plastic surgery\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6233674/v1/42a90d075b01b961871208ee.png"},{"id":80089841,"identity":"ccff4d0f-44f2-4187-ab85-91f2719171c3","added_by":"auto","created_at":"2025-04-07 18:24:46","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":36292,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eROC analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFig 4. \u003c/strong\u003eThe closer the ROC curve is to the upper left corner, the stronger the model's predictive ability; when the value approaches 1, the model's performance is better. The ROC curves for the training and testing sets are close to each other, indicating that the model performs well on both datasets and suggesting that there are no issues of overfitting or underfitting.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6233674/v1/f9205f84cdde1b09b5959bbd.png"},{"id":80089844,"identity":"59ce0d5d-3ac2-4fef-a823-881c20551ec2","added_by":"auto","created_at":"2025-04-07 18:24:46","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":394437,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFacial Information Collection and Processing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFig. 5：\u003c/strong\u003eStandardized outpatient 3D photographs, including frontal and lateral views (see A, B, C), were converted into 3D point cloud models via STL modeling to facilitate subsequent deep learning and feature point extraction.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6233674/v1/eade07d2011c515badb1de67.png"},{"id":80089285,"identity":"f791cab7-984a-4130-8aa6-6c45811737e2","added_by":"auto","created_at":"2025-04-07 18:08:46","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":126113,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of experimental results for accuracy, safety, and satisfaction.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6233674/v1/a2cd47e6368552738184aab8.png"},{"id":83830871,"identity":"722bdb5a-a885-4d7f-a1a3-c37212fc2e5a","added_by":"auto","created_at":"2025-06-03 11:39:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2698400,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6233674/v1/bc49d579-a807-4003-bdaa-55cc26215c45.pdf"},{"id":80089283,"identity":"d4c0cfa0-4892-4a12-97dd-a21b8f591722","added_by":"auto","created_at":"2025-04-07 18:08:46","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":21251,"visible":true,"origin":"","legend":"","description":"","filename":"completequestionnaire.docx","url":"https://assets-eu.researchsquare.com/files/rs-6233674/v1/faacd076e3e4fe8a0a7c378d.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Multi-Modal AI Plastic Surgery Diagnosis and Decision-Making System: Integration of Deep Learning and Psychological Assessment","fulltext":[{"header":"Introduction","content":"\u003cp\u003eFollowing the adjustment of China\u0026apos;s COVID-19 control policies, the macroeconomy entered a cyclical downturn phase (Q1 2023)(1, 2). Contact-intensive service industries (retail, dining, tourism) experienced operational disruptions and demand-side contractions, leading to a year-on-year decrease of 18.7% in industrial added value and resulting in a typical Keynesian effective demand deficiency phenomenon(3-5). This economic contraction was amplified through the multiplier effect of the \u0026quot;income-consumption-expectation\u0026quot; cycle, with the World Health Organization (WHO) standardizing mental health questionnaire (EQ-5D-5L) showing a 32.4% decline in urban residents\u0026apos; mental health index compared to pre-pandemic levels, while the anxiety disorder prevalence rate rose to 14.8%\u0026nbsp;(6-9). The impact of socio-economic issues on population mental health exhibits significant demographic heterogeneity: the youth group (18-24 years old) faces structural employment contradictions, with the 2023 college graduate survey indicating a 23% decrease in job supply index (PSI) compared to 2019 baseline values, while the job density index (JDI) increased by 17%, forming an expanding asymmetrical distribution of supply-demand gaps\u0026nbsp;(10-12). Meanwhile, the middle-aged group (35-50 years old) faces rigid career transition constraints, with National Bureau of Statistics tracking data showing that the median reemployment cycle for this group extended to 9.8 months (\u0026Delta;=+127%), significantly higher than the OECD average (\u0026Delta;=+62%) (OECD, 2023)\u0026nbsp;(13, 14).\u003c/p\u003e\n\u003cp\u003eIn this context, the medical aesthetics industry has shown unique development trends. Despite being widely regarded as a sunrise industry, industry data indicates that the market size contracted continuously from 2020 to 2022 (CAGR=-8.9%), and after policy easing in 2023, only a weak recovery of 1.2% was achieved, still failing to return to the 2019 baseline level (15-17). Notably, consumer decision-making mechanisms have undergone systemic changes. Based on a multi-factor regression analysis of 5,300 outpatient cases, our team observed that the acceptance of treatment plans by seekers of aesthetic services is influenced by a combination of factors including gender, age, psychological state, and price sensitivity, with some cases exhibiting cognitive biases in understanding indications\u0026nbsp;(18-20). In clinical practice, both physicians and patients face dual challenges in the decision-making process: on one hand, patients may not fully express their true needs, and on the other hand, the information physicians obtain during limited outpatient visits is often one-sided, making it difficult to comprehensively analyze patients\u0026apos; multi-dimensional needs. Therefore, developing tools that can analyze multi-source information and provide reasonable and safe treatment suggestions has become an important research direction to address these challenges.\u003c/p\u003e\n\u003cp\u003eIn response to this complex decision-making scenario, we decided to developed an intelligent assistance system based on multi-modal data fusion. The system integrates four-dimensional data sources:\u0026nbsp;①\u0026nbsp;high-precision 3D facial point cloud reconstruction model data (RMS=0.1mm);\u0026nbsp;②\u0026nbsp;standardized psychological assessment tools (EQ-5D-5L/PHQ-9/GAD-7);\u0026nbsp;③\u0026nbsp;dynamic economic affordability models;\u0026nbsp;④\u0026nbsp;social media interest and attention analysis. Through training using a self-supervised convolutional neural network framework (CAE-SCNN), our team\u0026apos;s previous research demonstrates that this algorithmic framework can complete relevant tasks with significantly reduced training samples while maintaining an 83.2% clinical recommendation compliance rate\u0026nbsp;(21, 22), showing a clear advantage over traditional deep learning methods (P\u0026lt;0.01)\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eA bibliometric analysis (Web of Science Core Collection 2020-2024) reveals a knowledge gap in this research area: among 2,438 relevant articles, only 31 related papers were found that involve multi-factor interaction analysis concerning AI, psychology, treatment recommendations, and plastic surgery, while only 6 studies integrate artificial intelligence with clinical decision-making. Furthermore, the field of multi-modal AI decision-making based on psychological, economic, and biological features remains a blank area in plastic surgery (the results of the bibliometric analysis are shown in Figure 1). This study constructs, for the first time, a decision support system that spans psychology, plastic surgery, and software engineering, providing data-driven transformation and upgrading solutions for the industry. The innovation of this research lies in integrating multi-modal data (3D facial point clouds, psychological assessment questionnaires, economic affordability models, and social media profiles) into an aesthetic treatment decision support system, addressing the gap in multi-modal AI decision-making based on psychological, economic, and biological features in plastic surgery. Additionally, through the innovative application of the self-supervised convolutional neural network framework (CAE-SCNN), The training results can still be good based on a smaller sample size., demonstrating a substantial improvement over traditional deep learning methods (P\u0026lt;0.01). This research provides new technical pathways for precision medicine in plastic surgery by combining artificial intelligence algorithms with psychological, economic, and biological features. In terms of plastic surgery development, this study has significant theoretical and practical implications. Theoretically, it expands the research boundaries of decision support systems in plastic surgery and offers a new research paradigm for the application of multi-modal data fusion and AI algorithms in medicine. Practically, the intelligent assistance system developed in this study enables clinicians to more comprehensively understand patient needs, optimize personalized treatment plans, and improve the efficiency and quality of medical services. Furthermore, by providing data-driven solutions for the digital transformation of the medical aesthetics industry, this research supports the industry\u0026apos;s transition from traditional models to intelligent and precise approaches, injecting new vitality into the sustainable development of plastic surgery.\u0026nbsp;\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003e1. Ethics \u0026amp; Privacy Protection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study strictly adhered to the principles outlined in the Declaration of Helsinki, China\u0026apos;s Ethical Review Measures for Biomedical Research Involving Humans, and the European Union\u0026apos;s General Data Protection Regulation (GDPR) to ensure ethical compliance throughout the research process and to maximize the protection of participants\u0026apos; rights. Specific measures are detailed below:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.1 Ethics approval:\u003c/strong\u003e Approved by the Ethics Committee of Xiangya Third Hospital (No. K23688) with independent monitoring;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.2 Informed consent:\u0026nbsp;\u003c/strong\u003eThree-level consent forms detailed data usage, anonymization, and \u0026nbsp; \u0026nbsp; \u0026nbsp; withdrawal rights; the facial photos used in this study and the collected questionnaire data have been obtained with informed consent from all participants, and the participant in Figure 5 has agreed to publish in an open access journal.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.3 Privacy measures\u003c/strong\u003e:Differential privacy (\u0026epsilon;=0.76, Laplace noise \u0026Delta;f=1.2) with k-anonymity (k=5) and l-diversity (l=2); AES-256 encrypted storage on private cloud with two-factor authentication; \u0026quot;Safe harbor\u0026quot; datasets containing only aggregated statistics were shared externally;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.4 Risk control\u003c/strong\u003e: Independent Data Monitoring Committee (IDMC) conducted quarterly audits to prevent unauthorized access.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2. Multimodal Database Construction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA multicenter cross-sectional study was conducted from December 1, 2023, to December 1, 2024, across 33 plastic surgery clinics in Changsha. Participants (n=5,543, response rate: 93.72%) were enrolled with the following data collected:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.1 Psychological assessment\u003c/strong\u003e: Standardized scales (EQ-5D-5L, PHQ-9, GAD-7, ISI, IES-R, HRQOL) and novel scales (COVID-19-related Plastic Surgery Intention Scale, Trauma Acceptance Scale);\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2 3D facial imaging\u003c/strong\u003e: Captured using iPhone 14 Pro under standardized conditions (object distance: 1.2m, aperture: f/2.8, depth-enhanced mode, dual-lens depth sensing);\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.3 Clinical data\u003c/strong\u003e: Physician recommendations, patient decisions, and postoperative outcomes (satisfaction, recovery time, re-visit rate) were automatically recorded via an electronic medical record (EMR) system.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3. Data Preprocessing and Modeling\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.1 Feature Engineering\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e3.1.1 Ordinal variables (e.g., insomnia severity): Encoded with ordinal labeling;\u003c/p\u003e\n\u003cp\u003e3.1.2 Nominal variables\u0026nbsp;(e.g., gender, occupation): Processed via one-hot encoding;\u003c/p\u003e\n\u003cp\u003e3.1.4 Continuous variables (e.g., age, income): Normalized using Z-score;\u003c/p\u003e\n\u003cp\u003e3.2.4 \u0026nbsp; Text data: Tokenized via Jieba and vectorized with TF-IDF for BiLSTM input.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 Self-supervised Multimodal Fusion\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA\u0026nbsp;two-phase training framework\u0026nbsp;was implemented to reduce annotation dependency:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2.1 Self-supervised pretraining\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003e① Contrastive learning (SimCLR variant) applied to unlabeled 3D point clouds and text data;\u003c/p\u003e\n\u003cp\u003e② Augmentations: Random rotation (\u0026plusmn;15\u0026deg;), occlusion (20% regions), and Gaussian noise (\u0026sigma;=0.01) for point clouds; synonym replacement (WordNet) for text;\u003c/p\u003e\n\u003cp\u003e③ Loss function: NT-Xent loss with negative pair margin set to 1.2.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2.2 Multimodal fusion\u003c/strong\u003e:\u003c/p\u003e\n\u003cp\u003e① Architecture: ResNet-50 for geometric features, BiLSTM for text, and FC-512 for clinical data;\u003c/p\u003e\n\u003cp\u003e② Cross-modal attention: 4-head attention dynamically weighted modality contributions;\u003c/p\u003e\n\u003cp\u003e③ Transfer learning: Frozen pretrained encoders reduced supervised data requirements by 63%.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2.3 Class Balancing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSMOTE-ENN hybrid sampling improved minority class (e.g., severe depression, n=227) F1-score from 0.68 to 0.81 (\u0026Delta;=+19.1%) while reducing overfitting risk by 22%.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3. 3D Point Cloud Reconstruction \u0026amp; QC\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e①\u0026nbsp;Point cloud generation: Agisoft Metashape (v2.0) produced dense clouds (mean reprojection error: 0.23mm, SD=0.07);\u003c/p\u003e\n\u003cp\u003e②\u0026nbsp;Illumination correction: Spherical harmonics decomposed diffuse/specular components;\u003c/p\u003e\n\u003cp\u003e③\u0026nbsp;Quality control: CloudCompare evaluated density (\u0026gt;500 pts/cm\u0026sup2;) and curvature continuity (Hausdorff distance \u0026lt;0.5mm), excluding outliers (n=43).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4. Model Training \u0026amp; Validation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e4.1 5-fold cross-validation: 5,543 cases were stratified (train/validation=8:2);\u003c/p\u003e\n\u003cp\u003e4.2 Loss function: Class-weighted cross-entropy (weights=1/\u0026radic;N);\u003c/p\u003e\n\u003cp\u003e4.3 Optimization: AdamW (lr=3e-5, \u0026beta;1=0.9, \u0026beta;2=0.999) with cosine annealing scheduler (T_max=50).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5. Clinical Double-blind Validation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5.1 Evaluation cohort\u003c/strong\u003e: 150 new patients independently assessed by 3 chief physicians (mean experience: 18.6 years) and the AI system;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5.2 Metrics\u003c/strong\u003e: Patient satisfaction: Likert 5-point scale (Cronbach\u0026apos;s \u0026alpha;=0.89); Physician ratings: Treatment accuracy (ICC=0.92) and safety (NCCN compliance \u0026ge;95%);\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistical Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData analysis was performed using SAS University Edition (v3.8), with a statistical significance threshold set at two-tailed \u0026alpha; = 0.05. All tests excluded multicollinearity interference through variance inflation factors (VIF \u0026lt; 5). The specific methods are as follows:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1. Between-group Comparative Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.1 Accuracy Comparison\u003c/strong\u003e: For the accuracy scores of treatment plans provided by physicians and the AI system (continuous variables), paired-sample t-tests were used to compare group differences, with the effect size quantified using Cohen\u0026apos;s d values (normality confirmed by Shapiro-Wilk test, p \u0026gt; 0.1).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.2 Safety Assessment\u003c/strong\u003e: For NCCN guideline compliance rates (binary variables), McNemar tests (paired chi-square tests) were applied to calculate odds ratios (OR) and 95% confidence intervals.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.3 Patient Satisfaction\u003c/strong\u003e: Based on a Likert 5-point scale (ordered categorical variable), Wilcoxon signed-rank tests were used to analyze group differences, and Hodges-Lehmann median offsets were reported.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2. Non-inferiority Test Design\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA non-inferiority margin (\u0026Delta; = 10%) was predefined. For the primary efficacy metrics (e.g., treatment accuracy) between the AI system and physicians, two-sided Z-tests were used to calculate 95% confidence intervals for group differences. Non-inferiority was established if the upper bound of the confidence interval was below \u0026Delta;.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3. Correlation and Predictive Modeling\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e3.1 Categorical Variable Associations: The relationships between gender (binary), education level (ordered categorical), and post-COVID-19 psychological metrics (continuous PHQ-9/GAD-7/EQ-5D-5L/PHQ-9/GAD-7/ISI/IES-R/HRQOL scores) were analyzed using:\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;3.2 Spearman rank correlation analysis (education level vs. psychological metrics). Mann-Whitney U tests (gender group comparisons).\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;3.3 Plastic Surgery Intention Prediction: A multivariate logistic regression model was constructed, with plastic surgery intention as the outcome variable. Covariates included age, gender, and anxiety/depression scores. Variables were selected using stepwise regression (AIC criterion), and model performance was evaluated using the area under the ROC curve (AUC).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4. Statistical Control and Correction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e4.1 Multiple Comparisons: Bonferroni corrections were applied to address multiple testing issues, adjusting the significance threshold to \u0026alpha;\u0026apos; = 0.05/k (k = number of tests).\u003c/p\u003e\n\u003cp\u003e4.2 Missing Data Handling: Missing values for continuous variables (\u0026lt;5%) were addressed using multiple imputation (PROC MI), while missing values for categorical variables were replaced by the mode.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5. Sensitivity Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e5.1 Robustness Verification: Bootstrap resampling (1,000 iterations) was used to validate the robustness of primary findings.\u003c/p\u003e\n\u003cp\u003e5.2 Non-normal Data: For non-normally distributed data (Kolmogorov-Smirnov test p \u0026lt; 0.05), robust regression (Huber-White standard errors) was applied in secondary analyses.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eThe overall process flow diagram can be seen in Figure 2.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1. Text data analysis (including questionnaires, diagnostic recommendations, and demographic baseline) and quality assessment of the text dataset.\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe statistical data on the participants\u0026apos; basic information is shown in Table 1; the statistical data on the participants\u0026apos; psychological state scores is shown in Table 2.\u003c/p\u003e\n\u003cp\u003eTable 1：participants\u0026apos; basic information\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"568\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 164px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eGender\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 132px;\"\u003e\n \u003cp\u003eEducation Level\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 117px;\"\u003e\n \u003cp\u003eIncome\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eGrouping\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003eYouth Group\u003c/p\u003e\n \u003cp\u003e（18-30）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eMiddle-Aged Group（30-60）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 48px;\"\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 51px;\"\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003eHighly Educated Group\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003eLow Education Group\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 59px;\"\u003e\n \u003cp\u003eHighly income gourp\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 59px;\"\u003e\n \u003cp\u003eLow income group\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003eNumber of People\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e3310\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e1885\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 48px;\"\u003e\n \u003cp\u003e879\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 51px;\"\u003e\n \u003cp\u003e4316\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e3810\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e1385\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 59px;\"\u003e\n \u003cp\u003e1681\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 59px;\"\u003e\n \u003cp\u003e3514\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 56px;\"\u003e\n \u003cp\u003ePercentage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e63.72%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e36.28%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 48px;\"\u003e\n \u003cp\u003e16.92%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 51px;\"\u003e\n \u003cp\u003e83.08%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e73.34%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e26.66%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 59px;\"\u003e\n \u003cp\u003e32.36%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 59px;\"\u003e\n \u003cp\u003e67.64%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTalbe2：participants\u0026apos; psychological state scores\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003ePopulation grouping\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003eEQ-5D-5L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003ePHQ-9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003eGAD-7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003eISI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003eIES-R\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003eHRQOL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eIntention for plastic surgery\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eYouth Group（mine\u0026plusmn;SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.772\u0026plusmn;0.825\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.456\u0026plusmn;1.791\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e1.112\u0026plusmn;1.811\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e1.238\u0026plusmn;1.690\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e17.042\u0026plusmn;9.484\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e5.899\u0026plusmn;4.253\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e7.765\u0026plusmn;0.477\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eMiddle-Aged Group（mine\u0026plusmn;SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.656\u0026plusmn;0.918\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.164\u0026plusmn;0.761\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e1.337\u0026plusmn;2.602\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e1.873\u0026plusmn;2.424\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e24.373\u0026plusmn;5.07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e10.019\u0026plusmn;4.601\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e6.629\u0026plusmn;1.184\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eP value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.033\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003emale（mine\u0026plusmn;SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.173\u0026plusmn;0.695\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.554\u0026plusmn;1.878\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.272\u0026plusmn;0.699\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.126\u0026plusmn;0.466\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e6.887\u0026plusmn;6.572\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e3.386\u0026plusmn;3.997\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e7.958\u0026plusmn;0.265\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003efemale（mine\u0026plusmn;SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.842\u0026plusmn;0.849\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.307\u0026plusmn;1.409\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e1.383\u0026plusmn;2.282\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e1.747\u0026plusmn;2.099\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e22.374\u0026plusmn;6.735\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e8.247\u0026plusmn;4.545\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e7.220\u0026plusmn;1.022\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eP value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.006\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eHighly education lever（mine\u0026plusmn;SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.743\u0026plusmn;0.849\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.379\u0026plusmn;1.577\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e1.128\u0026plusmn;1.997\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e1.372\u0026plusmn;1.873\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e19.397\u0026plusmn;9.041\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e7.381\u0026plusmn;4.872\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e7.422\u0026plusmn;0.922\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eLow education lever（mine\u0026plusmn;SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.635\u0026plusmn;0.945\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.136\u0026plusmn;0.795\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e1.646\u0026plusmn;2.896\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e2.154\u0026plusmn;2.707\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e22.191\u0026plusmn;7.164\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e7.719\u0026plusmn;4.398\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e6.819\u0026plusmn;1.169\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003eP value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 66px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eHighly income group（mine\u0026plusmn;SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e0.787\u0026plusmn;1.014\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.164\u0026plusmn;0.824\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e1.453\u0026plusmn;2.646\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e1.901\u0026plusmn;2.402\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e23.444\u0026plusmn;6.320\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e9.792\u0026plusmn;4.494\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e6.776\u0026plusmn;1.167\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eLow-income group（mine\u0026plusmn;SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e0.684\u0026plusmn;0.718\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.494\u0026plusmn;1.859\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.992\u0026plusmn;1.605\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e1.135\u0026plusmn;1.568\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e16.837\u0026plusmn;9.491\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e5.549\u0026plusmn;4.195\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e7.795\u0026plusmn;0.421\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eP value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 67px;\"\u003e\n \u003cp\u003e0.489\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003e0.000\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eThe analysis of questionnaire data mining revealed that\u003c/strong\u003e differences in anxiety, depression, and intention for cosmetic surgery after COVID-19 among different age groups (see Table 2 part 1). Specifically, the EQ-5D-5L health scale indicates that compared to the middle-aged group, the younger group has a poorer self-assessed health status; the results from the PHQ-9 depression screening scale show that the younger group has a higher severity of self-assessed depression compared to the middle-aged group; the Generalized Anxiety Disorder 7-item scale (GAD-7) indicates that the middle-aged group has higher anxiety levels than the younger group; the Insomnia Severity Index (ISI) shows that the middle-aged group experiences higher levels of insomnia than the younger group; the Impact of Event Scale-Revised (IES-R) demonstrates that the middle-aged group is significantly more affected by the epidemic event than the younger group; the Health-Related Quality of Life scale (HRQOL) reflects that due to COVID-19, the middle-aged group\u0026apos;s self-assessed health-related quality of life is poorer than that of the younger group; the intention for cosmetic surgery scale shows that the younger group has a higher negative intention towards cosmetic surgery than the middle-aged group, indicating that the middle-aged group has a higher intention for cosmetic surgery. All these differences have p-values less than 0.05, indicating statistical significance.\u003c/p\u003e\n\u003cp\u003eAdditionally, the study also found relationships between anxiety, depression, and intention for cosmetic surgery after COVID-19 with variables such as gender, education level, and income (see Table 2 part2，3，4). Specific details include differences in health status, depression severity, anxiety levels, insomnia severity, epidemic impact, and intention for cosmetic surgery among different gender, education level, and income groups, with most differences being statistically significant. It is worth noting that the p-value for the Health-Related Quality of Life scale (HRQOL) is greater than 0.05, indicating that the difference in income is not statistically significant, while differences among other variables are statistically significant.\u003c/p\u003e\n\u003cp\u003eWe further analyzed the relationships between psychological factors, gender, age, education, income, and their impact on the intention for plastic surgery decision-making in the post-pandemic era, as shown in Figure 3.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe correlation trends between anxiety, depression, and intention for plastic surgery with different age groups are shown in (Figure 3, A).\u003c/p\u003e\n\u003cp\u003eThe correlation coefficients between age and various factors at a 95% confidence interval are as follows:\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003eFor EQ-5D-5L, R = -0.05033, with a p-value of 0.0003.\u003c/li\u003e\n \u003cli\u003eFor PHQ9, R = -0.08541, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor GAD-7, R = 0.05986, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor ISI, R = 0.1563, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor IES-R, R = 0.3816, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor HRQol, R = 0.3182, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor Intention for plastic surgery, R = -0.5295, with a p-value less than 0.0001.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eAnxiety, depression, and intention for plastic surgery trends with different income levels are shown in (Figure 3, B).\u003c/p\u003e\n\u003cp\u003eThe correlation coefficients between income and various factors at a 95% confidence interval are as follows:\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003eFor EQ-5D-5L, R = 0.03392, with a p-value of 0.0145.\u003c/li\u003e\n \u003cli\u003eFor PHQ9, R = -0.09332, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor GAD-7, R = 0.1, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor ISI, R = 0.1892, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor IES-R, R = 0.4318, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor HRQol, R = 0.4302, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor Intention for plastic surgery, R = -0.677 with a p-value less than 0.0001.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe correlation trends between anxiety, depression, and intention for plastic surgery with different genders are shown in (Figure 3, C).\u003c/p\u003e\n\u003cp\u003eThe correlation coefficients between genders and various factors at a 95% confidence interval are as follows:\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003eFor EQ-5D-5L, R = 0.2907, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor PHQ9, R = -0.06165, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor GAD-7, R = 0.1945, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor ISI, R = 0.3012, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor IES-R, R = 0.6544, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor HRQol, R = 0.3783, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor Intention for plastic surgery, R = -0.2829 with a p-value less than 0.0001.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe correlation trends between anxiety, depression, and intention for plastic surgery with different educational backgrounds are shown in (Figure 3, D).\u003c/p\u003e\n\u003cp\u003eThe correlation coefficients between education lever and various factors at a 95% confidence interval are as follows:\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003eFor EQ-5D-5L, R = 0.0197, with a p-value =0.1558.\u003c/li\u003e\n \u003cli\u003eFor PHQ9, R = 0.02311, with a p-value =0.0959.\u003c/li\u003e\n \u003cli\u003eFor GAD-7, R = -0.00967, with a p-value=0.4860.\u003c/li\u003e\n \u003cli\u003eFor ISI, R = -0.0276, with a p-value=0.0744.\u003c/li\u003e\n \u003cli\u003eFor IES-R, R = 0.0344, with a p-value=0.0132.\u003c/li\u003e\n \u003cli\u003eFor HRQol, R = 0.1467, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor Intention for plastic surgery, R = -0.0519 with a p-value less than 0.0001.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe correlation trends between anxiety levels and intention for plastic surgery, as well as depression levels, are shown in (Figure 3, E).\u003c/p\u003e\n\u003cp\u003eThe correlation coefficients between anxiety levels and various factors at a 95% confidence interval are as follows:\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003eFor EQ-5D-5L, R = 0.0275, with a p-value =0.1349.\u003c/li\u003e\n \u003cli\u003eFor PHQ9, R = 0.03992, with a p-value =0.004.\u003c/li\u003e\n \u003cli\u003eFor ISI, R = 0.7265, with a p-value less than 0.0001.\u003c/li\u003e\n \u003cli\u003eFor IES-R, R = 0.1312, with a p-value less than 0.0001..\u003c/li\u003e\n \u003cli\u003eFor HRQol, R = 0.03366, with a p-value= 0.0153.\u003c/li\u003e\n \u003cli\u003eFor Intention for plastic surgery, R = -0.1608 with a p-value less than 0.0001.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eBased on the survey data analysis in the preceding text, we have clarified the significant correlation between psychological status and demographic characteristics in decision-making for plastic surgery. To further investigate this, we constructed a logistic regression model incorporating 12 predictive factors, systematically integrating the following dimensions:\u003c/p\u003e\n\u003col start=\"1\" type=\"1\"\u003e\n \u003cli\u003e\u003cstrong\u003eCore psychological scales\u003c/strong\u003e: Including depression (PHQ9), anxiety (GAD-7), insomnia (ISI), and post-traumatic stress (IES-R).\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eDemographic characteristics\u003c/strong\u003e: Covering age, gender, income, and education level.\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eHealth status indicators\u003c/strong\u003e: Including the health-related quality of life scale (HRQol) and the EuroQol 5-dimensional scale (EQ-5D-5L).\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eBased on the analysis of the correlations in the aforementioned questionnaire data, we can clearly determine that factors such as age, gender, education level, income, and various psychological dimensions significantly influence patients\u0026apos; intentions and decision-making regarding plastic surgery. Therefore, it is essential to incorporate data from patient psychological questionnaires and baseline surveys into the deep learning dataset when developing an AI-based plastic surgery recommendation system in order to construct a multidimensional predictive model.\u003c/p\u003e\n\u003cp\u003eDuring the model performance evaluation phase, we validated the predictive capabilities of the training and testing sets through ROC curve analysis. As shown in Figure 4, the AUC value for the training set reached 0.96, while the AUC value for the testing set was 0.95. The high similarity between these two curves indicates that the model demonstrates good stability and reliability. This result further confirms the superior performance of the multidimensional model in predicting patients\u0026apos; intentions regarding plastic surgery treatment decisions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2. \u0026nbsp;3D point cloud generation from facial photo data\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFacial information is the cornerstone of surgical treatment and consultation decisions. Therefore, we adopted standardized photography techniques to meticulously capture patient facial features and converted the 3D photographs into 3D point cloud data, providing a richer source of information for deep learning software. The standard photographs and point cloud models captured in our study are shown in Figure 5.\u003c/p\u003e\n\u003cp\u003eIn the outpatient setting, the standardized photography of patient facial images is shown in Figures 5 A, B, and C. We used Agisoft Metashape software to perform 3D scanning of the captured images, generating an STL 3D model, which was then subjected to topological optimization to ensure that the number of triangular facets remained below 50,000, followed by the creation of a dense point cloud. By downsampling the point cloud, we converted it into a standard point cloud model for deep learning in this study, and further validated the point cloud file using CloudCompare software. The specific steps included: ICP registration comparison (registration error \u0026le; 0.05 mm), outlier removal (threshold set to 3\u0026sigma; standard deviation), and normal consistency optimization (angle threshold of 15\u0026deg;). This process retained all 21 key facial landmark points, including the tip of the nose and inner canthus points.\u003c/p\u003e\n\u003cp\u003eUltimately, the point cloud model achieved a density of 40% (equivalent to a resolution of 0.2 mm), with a total of 19,300 effective data points (distribution density variance \u0026lt; 8%), and the model file size was compressed to 35% of the original data. Additionally, non-soft tissue interference points, such as hair (accounting for \u0026lt; 5%), were also removed (see Figure 5D).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3. Comparison of the AI Plastic Surgery Diagnostic and Treatment Recommendation System with Physician Diagnostic Recommendations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAfter completing the training of the AI plastic surgery diagnostic and recommendation system, we re-enrolled 150 outpatient patients, collecting patient questionnaires and facial information. We then conducted both AI-based plastic surgery diagnostics and recommendations and manual diagnostics and recommendations (provided by three senior chief physicians). Subsequently, we performed a double-blind evaluation of the diagnostic recommendations by a third-party physician, assessing aspects such as the accuracy and safety of the recommendations, as well as patient satisfaction with the diagnostic results. The comparative study and analysis of the results are detailed in Figure 6 and Tables 3, 4, and 5.\u003c/p\u003e\n\u003cp\u003eTable 3. Accuracy of treatment recommendations\u003c/p\u003e\n\u003ctable border=\"1\" cellpadding=\"0\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eproject\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAI group (n=150)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003edoctor group (n=150)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eNon inferiority test results\u0026ldquo;\u0026Delta;=0.2\u0026rdquo;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026zwnj;Mean\u003cstrong\u003e\u0026nbsp;(SD)\u003c/strong\u003e\u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.25 (0.16) \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.18 (0.22) \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026zwnj; \u003cstrong\u003ePaired t-test P-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.054 \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026zwnj;\u003cstrong\u003eCohen\u0026apos;s d effect quantity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.19 \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026zwnj; \u003cstrong\u003eNon inferiority Z-value \u0026amp; P-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eZ=2.12 \u0026amp; P=0.017 \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThe non-inferiority threshold \u0026Delta; is set at 0.2 points (based on clinical equivalence consensus) 23\u003c/p\u003e\n\u003cp\u003eConclusion: The accuracy of the AI group is not inferior to that of the physician group (P=0.017), and there is no statistically significant difference in the mean between the two groups (P=0.054)\u003c/p\u003e\n\u003cp\u003eTable 4. Comparison of Safety\u003c/p\u003e\n\u003ctable border=\"1\" cellpadding=\"0\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eProject\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAI group (n=150)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003edoctor group (n=150)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNon inferiority test results\u0026ldquo;\u0026Delta;=5%\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026zwnj;\u003cstrong\u003eSafety rate of diagnosis and treatment recommendations\u003c/strong\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e86.0% \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e88.7% \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026zwnj; \u003cstrong\u003eMcNemar test P-value\u003c/strong\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.124 \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026zwnj; \u003cstrong\u003eOdds ratio (OR)\u003c/strong\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.81 (95% CI: 0.58\u0026ndash;1.13) \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026zwnj; \u003cstrong\u003eNon inferiority Z-value/P-value\u003c/strong\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eZ=1.76/P=0.039 \u0026zwnj;23\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThe non-inferiority threshold \u0026Delta; is set at 5% (based on clinical safety consensus)\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTherefore the safety of the AI group was not inferior to that of the physician group (P=0.039), and the difference between the two groups was not statistically significant (P=0.124)\u003c/p\u003e\n\u003cp\u003eTable 5.\u0026nbsp;Patient satisfaction of treatment suggestion evaluation\u003c/p\u003e\n\u003ctable border=\"0\" cellpadding=\"0\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eProject\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;AI group (n=150)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;doctor group (n=150)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026zwnj;Mean (SD)\u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.21 (0.15) \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3.79 (0.25) \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026zwnj;Wilcoxon P value\u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026lt;0.001 \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026zwnj; Hodges Lehmann offset \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.42 \u0026zwnj;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026mdash;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eIn terms of the accuracy of treatment recommendations, the AI group had an average score of 4.25 (SD=0.16), slightly higher than the doctor group\u0026rsquo;s score of 4.18 (SD=0.22). The paired t-test results showed a P-value of 0.054, indicating that the difference between the two groups is not statistically significant. Furthermore, the non-inferiority test (Z=2.12, P=0.017, non-inferiority threshold \u0026Delta;=0.2) further confirmed that the accuracy of the AI group is not inferior to that of the doctor group. Therefore, we can conclude that the accuracy of the AI group is clinically equivalent to that of the doctor group, with no significant difference in the means between the two groups.\u003c/p\u003e\n\u003cp\u003eRegarding safety, the NCCN guideline compliance rate for the AI group was 86.0%, slightly lower than the 88.7% for the doctor group. The McNemar test yielded a P-value of 0.124, indicating that the difference between the two groups is not statistically significant. The non-inferiority test (Z=1.76, P=0.039, non-inferiority threshold \u0026Delta;=5%) showed that the safety of the AI group is not inferior to that of the doctor group. This indicates that the safety of the AI group is comparable to that of the doctor group, with no statistically significant difference between the two groups.\u003c/p\u003e\n\u003cp\u003eIn terms of patient satisfaction, the AI group had an average score of 4.21 (SD=0.15), significantly higher than the doctor group\u0026rsquo;s score of 3.79 (SD=0.25). The Wilcoxon test yielded a P-value of \u0026lt;0.001, indicating that the difference between the two groups is statistically significant. The Hodges-Lehmann offset was 0.42, further confirming that patient satisfaction in the AI group is significantly higher than in the doctor group, with a statistically significant difference.\u003c/p\u003e\n\u003cp\u003eIn summary, the AI plastic surgery diagnostic and recommendation system demonstrates excellent performance in terms of accuracy, safety, and patient satisfaction. Its accuracy and safety are comparable to those of the doctor group, while it significantly outperforms the doctor group in patient satisfaction. This indicates that the AI system has high clinical application value in plastic surgery diagnostic recommendations, providing patients with a superior service experience.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study validates the outstanding performance of an AI-based plastic surgery diagnostic recommendation system that integrates multi-modal data fusion in terms of accuracy, safety, and patient satisfaction. Comparative analysis with the physician group shows that the AI group performs comparably in treatment recommendation accuracy (mean difference not statistically significant, P=0.054) and safety (non-inferiority test passed, P=0.039), while significantly outperforming the physician group in patient satisfaction scores (P\u0026lt;0.001). These results indicate that the AI system holds high clinical application value in plastic surgery diagnostic recommendations, providing a potential for technological innovation in the field.\u003c/p\u003e\n\u003cp\u003eMultiple studies have shown that gender and income levels significantly influence beauty consumption behavior (22-24), suggesting that the development of plastic surgery diagnostic recommendation systems should differ from other intelligent diagnostic systems, particularly by incorporating psychological assessment factors as reference models. This study further reveals the moderating effects of age, education level, and psychological state on plastic surgery intentions and treatment decisions through multi-modal data analysis. Compared to research on appearance anxiety and social media related to plastic surgery decision-making (25-27)，this study constructs a more comprehensive and precise predictive model by integrating 3D facial imaging, psychological assessment data, and economic models. This innovative approach not only enriches the existing literature but also provides a new research perspective for the intelligent development of the plastic surgery field.\u003c/p\u003e\n\u003cp\u003eBy utilizing multi-modal data fusion (including 3D facial imaging, psychological assessment scales, and demographic models), this study successfully developed an efficient AI diagnostic recommendation system, pioneering a new technological pathway in plastic surgery. The introduction of the AI system not only helps physicians better understand patient needs but also optimizes personalized treatment plans, enhancing the efficiency and quality of medical services. The findings provide practical guidance for the digital transformation of the plastic surgery industry, specifically in the following areas:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1. Adapting to Changing Customer Perceptions\u003c/strong\u003e\u003cstrong\u003e:\u003c/strong\u003e In the post-pandemic era, emphasizing health and natural beauty through transparent education and social media can enhance patient trust.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2. Addressing Price Sensitivity: Adjusting treatment strategies based on patients\u0026apos; economic income can attract budget-conscious consumers during economic downturns.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3. Promoting Remote Consultation Services\u003c/strong\u003e\u003cstrong\u003e:\u003c/strong\u003e Utilizing AI technology for contactless consultations can enhance patient acceptance and safety through online consultations and simulation technologies.\u003c/p\u003e\n\u003ch4\u003eLimitations and Future Directions\u003c/h4\u003e\n\u003cp\u003eDespite the significant progress made in this study, several limitations remain:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.Sample Limitations\u003c/strong\u003e: The study\u0026apos;s sample is primarily concentrated in specific regions, which may limit the generalizability of the results. Future research should expand the sample range to include diverse geographic and cultural backgrounds to validate the robustness of the model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.Model Generalization\u003c/strong\u003e: The current model\u0026apos;s training data is based on a specific population; future efforts should focus on optimizing algorithms to adapt to diverse clinical scenarios.\u003c/p\u003e\n\u003cp\u003e3.Long-term Effect Evaluation: This study only conducted a short-term observation. and has not assessed the sustained impact of the AI system in long-term clinical applications or the dynamic changes in patient satisfaction.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTo further optimize the research, future efforts could focus on:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.Expanding Sample Size\u003c/strong\u003e: Increasing the sample size and geographic coverage to validate the robustness and applicability of the model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.Optimizing Algorithm Framework\u003c/strong\u003e\u003cstrong\u003e:\u003c/strong\u003e Exploring more advanced AI algorithms, such as deep learning and transfer learning techniques, to further enhance the model\u0026apos;s predictive accuracy and generalization capabilities(28, 29).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3. Longitudinal Follow-up Studies\u003c/strong\u003e\u003cstrong\u003e:\u003c/strong\u003e Conducting follow-up studies to evaluate the long-term effects of the AI system in clinical applications, including patient satisfaction, treatment outcome stability, and the sustained impact on healthcare efficiency (30).\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study successfully constructed an efficient AI plastic surgery diagnostic recommendation system through multi-modal data fusion and a self-supervised learning framework, providing innovative solutions for the digital transformation of the plastic surgery field. The results indicate that the AI system significantly enhances diagnostic accuracy, optimizes treatment plans, and improves patient experience. Future research will further refine the technical framework and expand its application scope, supporting the sustainable development of the plastic surgery industry while offering patients more efficient and personalized medical services.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eInformed Consent Statement\u003c/h2\u003e\n\u003cp\u003eThe facial photos used in this study and the collected questionnaire data have been obtained with informed consent from all participants, and the participant in Figure 5 has agreed to publish in an open access journal.\u003c/p\u003e\n\u003ch2\u003eConflict of interest statement\u003c/h2\u003e\n\u003cp\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eXiaohui Qiu: Conceptualized the study, designed the research framework, and led the integration of deep learning with psychological assessment in the multi-modal AI system. Supervised the entire project and finalized the manuscript.Shuyue Chen: Conducted psychological assessments, collected and analyzed clinical data, and contributed to the interpretation of psychological evaluation results to ensure the system\u0026apos;s applicability in patient care.Jiangjie Tang: Designed and implemented the biological experiments, including wound healing assays and histological analysis, ensuring the accuracy and reliability of biological data.Qiuyang Chen: Developed and optimized the deep learning algorithms, focusing on model training and validation to ensure the system\u0026apos;s diagnostic accuracy and decision-making efficiency.Shenghui Liao: Integrated the AI components with psychological assessment tools, developed the user interface, and ensured the system\u0026apos;s functionality and user-friendliness.Jianda Zhou: Conceptualized the study, designed the research framework, and led the integration of deep learning with psychological assessment in the multi-modal AI system. Supervised the entire project and finalized the manuscript.All authors reviewed and approved the final version of the manuscript.\u003c/p\u003e\n\u003ch2\u003eAcknowledgement\u003c/h2\u003e\n\u003cp\u003eThe authors greatly acknowledge the financial support from the National Natural Science Foundation of China (No. 52305313, No .52375341, No.81872219), FuRong Laboratory, Changsha 410078, Hunan, China（No.2023SK2115).\u003c/p\u003e\n\u003ch2\u003eData Availability\u003c/h2\u003e\n\u003cp\u003eThe data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWu J, Zhan X, Xu H, Ma C (2023): The economic impacts of COVID-19 and city lockdown: Early evidence from China. \u003cem\u003eStructural change and economic dynamics\u003c/em\u003e. 65:151-165.\u003c/li\u003e\n\u003cli\u003eShen Y, Sun A, Zhou Z, Jia D (2024): Digital finance and wealth inequality: Evidence from a big tech platform in China during the COVID-19 pandemic. \u003cem\u003ePacific-Basin Finance Journal\u003c/em\u003e. 83:102226.\u003c/li\u003e\n\u003cli\u003eJean-fran\u0026ccedil;ois Piluso N (2025): The Critical Mistake of New Keynesian Economics, Demonstrated in a Revised Model of Blanchard and Kiyotaki. \u003cem\u003eReview of Political Economy\u003c/em\u003e.1-16.\u003c/li\u003e\n\u003cli\u003eKiviruusu O, Ranta K, Lindgren M, Haravuori H, Sil\u0026eacute;n Y, Therman S, et al. (2024): Mental health after the COVID-19 pandemic among Finnish youth: a repeated, cross-sectional, population-based study. \u003cem\u003eThe Lancet Psychiatry\u003c/em\u003e. 11:451-460.\u003c/li\u003e\n\u003cli\u003eBistagnino F, Subramanian A, Tovani-Palone MR (2025): Navigating the (Post-) Pandemic Landscape: An Analysis of COVID-19\u0026rsquo;s Current Status and Future Implications. \u003cem\u003eDisaster Medicine and Public Health Preparedness\u003c/em\u003e. 19:e44.\u003c/li\u003e\n\u003cli\u003eJutakeo J, Kuptniratsaikul V, Limampai P, Akewanlop C, Hantaweepant C, Thephamongkhol K, et al. A Preliminary Report of the Effects of Dharma Creative Art Therapy on Psychological Impacts and Quality of Life of Thai Cancer Patients: A Non-randomized Trial.\u003c/li\u003e\n\u003cli\u003eMachrumnizar M, Muyana N, Bachtiar A, Kusumaratna RK, Suyanto J (2023): Impact of co-epidemic tuberculosis and diabetes mellitus on health-related quality of life: A review. \u003cem\u003eJurnal Aisyah: Jurnal Ilmu Kesehatan\u003c/em\u003e. 8.\u003c/li\u003e\n\u003cli\u003eWang G, Sabran K (2024): Assessing depression and anxiety among young adults after epidemics and pandemics: a cross-sectional study in Anyang, China. \u003cem\u003eScientific Reports\u003c/em\u003e. 14:2759.\u003c/li\u003e\n\u003cli\u003eHall BJ, Li G, Chen W, Shelley D, Tang W (2023): Prevalence of depression, anxiety, and suicidal ideation during the Shanghai 2022 Lockdown: A cross-sectional study. \u003cem\u003eJournal of affective disorders\u003c/em\u003e. 330:283-290.\u003c/li\u003e\n\u003cli\u003eMei DW (2023): New changes and challenges in the youth employment in China after COVID-19. \u003cem\u003eAsian Soc Sci\u003c/em\u003e. 19:84.\u003c/li\u003e\n\u003cli\u003eYu Z, Liu L, Zhang X (2024): Bridging the gap: Enhancing employment opportunities for Normal University graduates in China\u0026rsquo;s knowledge economy. \u003cem\u003eJournal of the Knowledge Economy\u003c/em\u003e.1-38.\u003c/li\u003e\n\u003cli\u003eZheng S, Yan Y (2024): Changes in employment psychology of Chinese university students during the two stages of COVID-19 control and their impacts on their employment intentions. \u003cem\u003eFrontiers in Psychology\u003c/em\u003e. 15:1447103.\u003c/li\u003e\n\u003cli\u003eAyhan F, Elal O (2023): The IMPACTS of technological change on employment: Evidence from OECD countries with panel data analysis. \u003cem\u003eTechnological Forecasting and Social Change\u003c/em\u003e. 190:122439.\u003c/li\u003e\n\u003cli\u003eCruz MD (2023): Labor productivity, real wages, and employment in OECD economies. \u003cem\u003eStructural Change and Economic Dynamics\u003c/em\u003e. 66:367-382.\u003c/li\u003e\n\u003cli\u003eLi Z, Li T, Zhang L, Zhu Y, Yu Z, Song B (2024): Impact of Public Health and Social Measures on Cosmetic Treatments in the COVID‐19 Pandemic: A Retrospective Multi‐Center Study Combined With a Questionnaire‐Based Cross‐Sectional Study. \u003cem\u003eJournal of Cosmetic Dermatology\u003c/em\u003e. 23:3800-3808.\u003c/li\u003e\n\u003cli\u003eYoon S, Kim YA (2024): Aesthetic Plastic Surgery Issues During the COVID-19 Period Using Topic Modeling. \u003cem\u003eInternational Journal on Advanced Science, Engineering \u0026amp; Information Technology\u003c/em\u003e. 14.\u003c/li\u003e\n\u003cli\u003eQiao Z, Deng Y, Wang X, Sun Y, Xiong X, Meng X, et al. (2023): The impact of COVID-19 on plastic and reconstructive surgery in China: A single-centre retrospective study. \u003cem\u003eJournal of Plastic, Reconstructive \u0026amp; Aesthetic Surgery\u003c/em\u003e. 76:160-168.\u003c/li\u003e\n\u003cli\u003eZhang Y, Jiang M, Liu J, Liu B (2024): Pursuing beauty: socio-cultural and labor-economic determinants of cosmetic surgery consideration among female college students in China. \u003cem\u003eBMC psychology\u003c/em\u003e. 12:519.\u003c/li\u003e\n\u003cli\u003eMohammed DI, Ibrahim RH (2023): Exploring the impact of psychological factors on cosmetic surgery acceptance: A cross-sectional study. \u003cem\u003eInformatics in Medicine Unlocked\u003c/em\u003e. 39:101231.\u003c/li\u003e\n\u003cli\u003eRahman E, Rao P, Webb WR, Garcia PE, Ioannidis S, Tam E, et al. (2024): Integrating Psychological Insights into Aesthetic Medicine: A Cross-Generational Analysis of Patient Archetypes (IMPACT Study). \u003cem\u003eAesthetic Plastic Surgery\u003c/em\u003e.1-15.\u003c/li\u003e\n\u003cli\u003eQiu X, Zhong C, Chen Q, Zhao Y, Yang T, Zhou J, et al. (2025): Zygomatic Osteotomy surgery design software based on skull CT scans-Self-supervised algo reduces workload. \u003cem\u003eJournal of Cranio-Maxillofacial Surgery\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eQiu X, Han W, Dai L, Zhang Y, Zhang J, Chai G, et al. (2022): Assessment of an Artificial Intelligence Mandibular Osteotomy Design System: A Retrospective Study. \u003cem\u003eAesthetic Plastic Surgery\u003c/em\u003e. 46:1303-1313.\u003c/li\u003e\n\u003cli\u003eLi Z, Cao H, Yu N, Qin F, Li Y, Li Z, et al. (2023): What factors will influence patients when choosing plastic surgeons? A behaviour analysis of Chinese patients. \u003cem\u003eJournal of Plastic, Reconstructive \u0026amp; Aesthetic Surgery\u003c/em\u003e. 83:57-68.\u003c/li\u003e\n\u003cli\u003eChou DW, Layfield E, Prasad K, Shih C, Brandstetter K (2023): Gender and ethnic diversity in academic facial plastic surgery. \u003cem\u003eThe Laryngoscope\u003c/em\u003e. 133:1869-1874.\u003c/li\u003e\n\u003cli\u003eMironica A, Popescu CA, George D, Tegzeșiu AM, Gherman CD (2024): Social media influence on body image and cosmetic surgery considerations: a systematic review. \u003cem\u003eCureus\u003c/em\u003e. 16.\u003c/li\u003e\n\u003cli\u003eK\u0026Uuml;\u0026Ccedil;\u0026Uuml;K \u0026Ouml;ZT\u0026Uuml;RK G, BAŞER E, \u0026Ouml;ZOCAK H (2023): Investigation of Basic Personality Traits and Social Appearance Anxiety of Individuals Undergoing Aesthetic Surgery Operations: A Descriptive Study. \u003cem\u003eTurkiye Klinikleri Journal of Nursing Sciences\u003c/em\u003e. 15.\u003c/li\u003e\n\u003cli\u003eTokg\u0026ouml;z E, Carro MA (2023): Engineering psychology of facial plastic surgery patients. \u003cem\u003eCosmetic and reconstructive facial plastic surgery: A review of medical and biomedical engineering and science concepts\u003c/em\u003e: Springer, pp 343-365.\u003c/li\u003e\n\u003cli\u003eChen J, Huang R, Chen Z, Mao W, Li W (2023): Transfer learning algorithms for bearing remaining useful life prediction: A comprehensive review from an industrial application perspective. \u003cem\u003eMechanical Systems and Signal Processing\u003c/em\u003e. 193:110239.\u003c/li\u003e\n\u003cli\u003eHakim M, Omran AAB, Ahmed AN, Al-Waily M, Abdellatif A (2023): A systematic review of rolling bearing fault diagnoses based on deep learning and transfer learning: Taxonomy, overview, application, open challenges, weaknesses and recommendations. \u003cem\u003eAin Shams Engineering Journal\u003c/em\u003e. 14:101945.\u003c/li\u003e\n\u003cli\u003eYan P, Guo J, Su X, Bai C (2024): Long-Term Tracking of Evasive Urban Target Based on Intention Inference and Deep Reinforcement Learning. \u003cem\u003eIEEE Trans Neural Netw Learn Syst\u003c/em\u003e. 35:16886-16900.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Artificial intelligence, COVID-19 epidemic, plastic surgery, education, income, gender, mental state","lastPublishedDoi":"10.21203/rs.3.rs-6233674/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6233674/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003ePurpose\u003c/strong\u003e: To design and develop an AI-based plastic surgery recommendation system using 3D photographs and psychological questionnaire surveys, aiming to provide personalized treatment solutions for the plastic surgery industry.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e: Based on artificial intelligence technology, this study utilized patients' 3D photographs and psychological questionnaire results as training samples to construct a personalized AI-based plastic surgery recommendation system. This system comprehensively considers factors such as patients' anxiety levels, economic status, and psychological expectations. The study selected 5,543 cases of plastic surgery outpatients aged 18 to 55 years, collected their 3D photographs and questionnaire data, and used these for AI system training. The software predicted treatment projects and compared them with doctors' predictions to validate the system's accuracy and patient satisfaction. Third-party doctors evaluated the system's safety, ultimately developing an efficient and accurate plastic surgery recommendation system.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e: The economic downturn in the post-COVID-19 era significantly impacted psychological health and the plastic surgery industry. Factors such as age, education level, income, and gender had significant effects on patients' psychological state and treatment willingness. The AI system integrated patients' psychological state, gender, income, and physical characteristics, providing personalized plastic surgery treatment suggestions and achieving a 93.25% patient satisfaction rate.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion\u003c/strong\u003e: The AI-based personalized plastic surgery recommendation system offers an innovative solution for the industry, enhancing the accuracy of treatment suggestions and patient satisfaction, thereby promoting sustainable development. In the post-pandemic era, the plastic surgery industry should focus on patients' physical, psychological, and economic factors to achieve personalized services.\u003c/p\u003e","manuscriptTitle":"Multi-Modal AI Plastic Surgery Diagnosis and Decision-Making System: Integration of Deep Learning and Psychological Assessment","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-07 18:08:41","doi":"10.21203/rs.3.rs-6233674/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"98365d59-a22f-4806-abc4-52ba157d758f","owner":[],"postedDate":"April 7th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":46449270,"name":"Biological sciences/Biological techniques/Imaging"},{"id":46449271,"name":"Biological sciences/Biological techniques/Software"}],"tags":[],"updatedAt":"2025-06-03T11:39:40+00:00","versionOfRecord":[],"versionCreatedAt":"2025-04-07 18:08:41","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6233674","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6233674","identity":"rs-6233674","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}