Scientific Abstract Presentations: Research & Technology Session 6 Saturday, October 17 Sat, Oct 17 2:00-3:30 p.m. CDT

Purpose: Melanoma is a growing global public health concern, with projected increases in incidence (1,2). In Brazil, marked socioeconomic and demographic heterogeneity may contribute to regional differences in disease burden (3-5). This study characterized melanoma epidemiology in Brazil from 2019 to 2023, focusing on regional variation in incidence, mortality, treatment access, and temporal trends.

Methods: A cross-sectional ecological time-series study was conducted using DATASUS (SIH, SIA, SIM) and INCA data from January 2019 to December 2023. Melanoma cases were identified by ICD-10 codes. Variables included incidence, mortality, age, sex, region (North, Northeast, Central-West, Southeast, South), hospital costs, and treatment location. Age-standardized rates were calculated using the Segi (1960) standard population. ARIMA and Holt-Winters models, negative binomial regression (population as offset), and hierarchical clustering were applied. Significance level was 5%.

Results: Between 2019 and 2023, 27,827 cases were recorded (mean annual incidence: 2.65/100,000). The South had the highest age-standardized incidence and formed a distinct high-incidence cluster. Incidence peaked around 60 years in all regions, with no significant annual percent change. A total of 9,739 deaths occurred (mean annual mortality: 0.93/100,000). Although the Southeast had the highest absolute number of deaths, the South showed the highest mortality rate (2.09/100,000). Mortality was higher among males and individuals ≥66 years. Age-standardized mortality remained stable nationwide. While 99% of patients were treated within their region of residence, marked disparities in time-to-treatment were observed. Southern states had shorter intervals (as low as 48 days in Paraná), whereas Northern states showed extreme delays (up to 683 days in Roraima). Migration to referral centers in the South/Southeast was associated with shorter waiting times. Seasonal incidence patterns were observed, without sustained national trends.

Conclusion: Melanoma in Brazil shows pronounced regional disparities. The South bears the highest incidence and mortality, whereas Northern states face substantial treatment delays. Despite global projections of rising incidence, national age-standardized rates remained stable from 2019–2023. Reducing regional inequalities in oncologic infrastructure is essential for improving equitable melanoma care within Brazil's Unified Health System.

References: 1- Arnold M, Singh D, Laversanne M, Vignat J, Vaccarella S, Meheus F, et al. Global Burden of Cutaneous Melanoma in 2020 and Projections to 2040. JAMA Dermatology [Internet]. 2022 Mar 30;158(5):495–503. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8968696/ 2- Erdmann F, Lortet-Tieulent J, Schüz J, Zeeb H, Greinert R, Breitbart EW, et al. International trends in the incidence of malignant melanoma 1953-2008-are recent generations at higher or lower risk? International Journal of Cancer. 2012 May 21;132(2):385–400. 3- de Melo AC, Wainstein AJA, Buzaid AC, Thuler LCS. Melanoma signature in Brazil: epidemiology, incidence, mortality, and trend lessons from a continental mixed population country in the past 15 years. Melanoma Research. 2018 Dec;28(6):629–36. ‌4- Institute for Health Metrics and Evaluation. GBD Compare [Internet]. VizHub. Institute for Health Metrics and Evaluation; 2021. Available from: https://vizhub.healthdata.org/gbd-compare/ 5- Souza RJSP de, Mattedi AP, Rezende ML, Corrêa M de P, Duarte EM. Estimativa do custo do tratamento de câncer de pele tipo melanoma no Estado de São Paulo - Brasil. Anais Brasileiros de Dermatologia. 2009 Jul;84(3):237–43.

PURPOSE: Vascularized composite allotransplantation (VCA) enables complex reconstruction but remains limited by a 4 to 6-hour ex vivo preservation window under static cold storage in University of Wisconsin (UW) solution. This window, defined by progressive loss of tissue integrity, restricts the donor pool, imposes surgical time constraints, and may compromise graft outcomes. We investigated whether supplementing UW with adipose-derived stromal cell-conditioned media (UW-CM) improves graft viability during ex vivo cold storage in a rat osteomyocutaneous flap model. Extending the viability window could expand the donor pool, increase operative flexibility, and limit postoperative graft complications.

METHODS: We stored rat femoral vessel-based osteomyocutaneous grafts (superficial epigastric skin flap, quadriceps muscle, femur) ex vivo in static cold storage in saline (SAL), UW, or UW-CM and performed analyses at 6 and 48 hours. We developed a histopathological injury severity score (HISS) for manual visual assessment of muscle viability using digital H&E images, scoring muscle fiber necrosis (1=normal, 2=mild/moderate necrosis, 3=severe necrosis) between two reviewers. We also trained HALO® AI to quantify the necrotic-to-intact muscle area ratio. We measured cytochrome C concentration by ELISA for apoptosis (n=3 wells per sample ). We compared HISS with Kruskal-Wallis and analyzed necrotic area and cytochrome C with two-way ANOVA.

RESULTS: Manual HISS scores (median [IQR]) did not differ across preservation solutions overall (p=0.3376) or at either time point (6 hours: SAL n=5, 2.5 [1.6–2.8]; UW n=5, 1.5 [1.4–2.0]; UW-CM n=2, 1.9 [1.75–2.0], p=0.225; 48 hours: SAL n=5, 2.0 [2.0–3.0]; UW n=6, 2.25 [2.0–3.0]; UW-CM n=5, 2.0 [1.75–2.5], p=0.534). AI-based histological pattern analysis demonstrated a significant effect of preservation solution on the necrotic-to-intact muscle area ratio (p=0.006) and a significant time × solution interaction (p=0.0395), indicating differential progression of muscle necrosis from 6 to 48 hours across groups. Necrotic-to-intact muscle area did not differ at 6 hours, but at 48 hours UW and UW-CM reduced muscle necrosis compared to saline by 66% and 55%, respectively (p=0.003 and p=0.031). UW and UW-CM did not have statistically significant difference at either time point (6 hours: p=0.5697; 48 hours: p=0.387). Cytochrome C concentrations were higher at 48 hours compared to 6 hours regardless of preservation solution (p=0.002), with no differences between groups (p=0.815), suggesting mechanisms beyond classical apoptosis may contribute to injury at these time points.

CONCLUSIONS: At 48 hours of static cold storage, UW and UW-CM reduced muscle necrosis compared with saline, with distinct trajectories of injury over time. Manual HISS did not detect these differences, whereas AI-based morphometric analysis identified solution-dependent effects. These findings support further refinement of UW-CM as a cell-free adjunct to standard VCA preservation. Beyond extended ex vivo preservation, this approach may have broader implications in mitigating ischemic injury in free tissue transfer and other reconstructive settings, including cases complicated by vascular comorbidities. By potentially improving tissue resilience during ischemic intervals, UW-CM supplementation may offer greater flexibility and more predictable graft outcomes beyond VCA.

Introduction: Delayed epithelialization of meshed skin grafts remains a major challenge in the treatment of full-thickness burns, particularly when limited donor skin necessitates high mesh ratios. Prolonged closure of mesh interstices increases the risk of infection, worsens patient outcomes, and increases healthcare costs. Keratinocyte migration is critical for epithelialization and is negatively regulated by β-adrenergic signaling [1-3]. Beta-blockers have been shown to enhance keratinocyte migration and improve wound healing in experimental models [2,4]; however, their efficacy in full-thickness burns reconstructed with meshed skin grafts has not been evaluated in a clinically relevant large-animal model. We therefore investigated the safety and efficacy of topical timolol in an ovine model of meshed autologous skin graft reconstruction following full-thickness burns.

Methods: Adult female Merino sheep underwent creation of six 25 cm² full-thickness flame burns on the dorsum. After 24 hours, eschars were excised and reconstructed with 4:1 meshed split-thickness autologous skin grafts (Day 0). Wounds were paired contralaterally and randomized to receive either topical timolol 0.5% (0.02 mL/cm²) or saline control, applied daily through Day 14. The epithelialization rates on Day 12, 13, and 14 were determined by planimetric analysis and compared using paired t-tests. Additionally, times to reach 85%, 90%, and 95% epithelialization were compared using survival time analysis with the log-rank test. Wound perfusion was measured using laser Doppler imaging and analyzed using two-way repeated-measures ANOVA followed by Bonferroni correction. Systemic hemodynamics and blood glucose levels were monitored to assess safety. At Day 14, histologic analysis and RT-qPCR were performed to evaluate epidermal thickness, inflammatory cell infiltration, angiogenesis, and expression of genes related to keratinocyte differentiation, TGFβ signaling, epithelial–mesenchymal transition, and myofibroblast activation. Histologic parameters and gene expression levels were compared using paired t-tests.

Results: All grafts healed without infection or hematoma. At Day 12, 13, and 14, epithelialization was significantly greater in the timolol-treated wounds compared with controls (68.88 ± 7.30% vs. 56.31 ± 6.52%, P = 0.0169 at Day 12, 76.84 ± 6.71% vs. 66.12 ± 6.06%, P = 0.0487 at Day 13, 84.46 ± 5.89% vs. 74.30 ± 5.86%, P = 0.0158 at Day 14). Time to reach 85%, 90%, and 95% epithelialization was significantly shorter in the timolol group (P = 0.0354, 0.0104, and 0.0313, respectively). Laser Doppler imaging demonstrated no differences in wound perfusion between groups at any time point. No systemic adverse effects were observed, including hypoglycemia, hypotension, or bradycardia. Histologically, epidermal thickness was greater in the timolol group (213.1 ± 15.6 µm vs. 183.5 ± 16.3 µm, P = 0.0203). Inflammatory cell counts (neutrophils, macrophages) and vessel density were similar between groups. RT-qPCR revealed no significant differences in transcriptional markers of proliferation (PCNA), keratinocyte differentiation (KRT1, KRT10), TGFβ signaling (TGFβ1, TGFβR1/2), epithelial–mesenchymal transition (VIM, FN1), or myofibroblast activation (COL1A1, COL3A1, ACTA2, MMP1).

Conclusions: Topical timolol significantly accelerates epithelialization of meshed autologous skin grafts in full-thickness burn wounds in a clinically relevant ovine model. The effect appears independent of changes in wound perfusion, inflammation, or transcriptional activation of TGFβ-, EMT-, or myofibroblast-related pathways, suggesting a mechanism distinct from angiogenic or proliferative modulation. Importantly, no systemic adverse effects were observed. Given its established safety profile, low cost, and widespread clinical availability, topical β-adrenergic blockade represents a promising adjunct to improve graft healing and potentially reduce complications associated with delayed epithelialization in burn reconstruction.

References: (1) Jia S, Wang X, Wang G, Wang X. Mechanism and application of β-adrenoceptor blockers in soft tissue wound healing. Med Res Rev. 2024;44(1):422-452. doi:10.1002/med.21984 (2) Sivamani RK, Pullar CE, Manabat-Hidalgo CG, et al. Stress-mediated increases in systemic and local epinephrine impair skin wound healing: potential new indication for beta blockers. PLoS Med. 2009;6(1):e12. doi:10.1371/journal.pmed.1000012 (3) Sarsik SM, El-Amawy HS. Uses of eye drops in dermatology, literature review. J Dermatolog Treat. 2022;33(6):2758-2770. doi:10.1080/09546634.2022.2079598 (4) Freiha M, Achim M, Gheban BA, Moldovan R, Filip GA. In Vivo Study of the Effects of Propranolol, Timolol, and Minoxidil on Burn Wound Healing in Wistar Rats. J Burn Care Res. 2023;44(6):1466-1477. doi:10.1093/jbcr/irad057

Purpose: Adipose tissue exhibits biological heterogeneity, yet the relative contributions of systemic factors such as BMI and age versus intrinsic depot-specific properties remain unclear. Prior work has largely focused on obesity-related changes in subcutaneous and visceral depots, but how adipocyte morphology and extracellular matrix (ECM) composition vary across anatomical sites and whether these structural differences are systemically driven or locally regulated is poorly understood. Importantly, distinguishing these drivers is critical for understanding depot-specific behavior in aging, metabolic disease, and reconstructive applications. This study aimed to distinguish systemic from depot-specific determinants of adipose microarchitecture by analyzing adipocyte size and ECM fraction across multiple tissue sources while adjusting for BMI and age.

Methods: Adipose tissue was collected from patients undergoing breast reduction, mastectomy, abdominoplasty, blepharoplasty, and lipoma excision (n=234). Histologic sections were quantified for cross-sectional adipocyte area (µm²) and ECM fraction (matrix %) using ImageJ. Patient demographics, including BMI, age and sex were recorded. Multivariable ANCOVA was used to assess the independent effects of systemic factors and tissue source. Interaction terms (age × depot) were evaluated to determine whether aging-related remodeling differed by anatomical site.

Results: Adipocyte size was predominantly systemically regulated. BMI demonstrated the strongest independent association (β=35.9 µm² per BMI unit, p<0.001), and age also contributed significantly (β=6.8 µm² per year, p=0.009). After adjustment for BMI and age, tissue source did not significantly influence adipocyte size, indicating that hypertrophy is largely driven by systemic metabolic factors rather than intrinsic depot identity. In contrast, ECM fraction was strongly depot-specific (overall R²=0.425, p<0.001) and independent of BMI and age. Periorbital adipose samples collected from blepharoplasty exhibited the highest matrix fraction (8.14 ± 3.19%), whereas breast reduction (3.71 ± 1.33%) and abdominal depots (3.96 ± 1.29%) demonstrated substantially lower ECM content. Mastectomy tissue showed intermediate but more variable ECM levels (5.33 ± 3.63%). Age alone did not significantly affect ECM composition. Interaction analysis revealed that aging-related ECM remodeling occurred selectively in breast tissue collected from mastectomy (age × depot, p=0.032), whereas periorbital, breast and abdominal depots remained structurally stable across age groups. Sex was not an independent determinant of adipocyte size or ECM fraction.

Conclusions: Adipocyte hypertrophy is primarily systemically driven by BMI and age, whereas ECM composition reflects intrinsic depot-specific structural identity. Most depots demonstrate remarkable ECM stability across the age span, with selective age-associated ECM decreased fraction observed only in oncologic breast tissue. These findings delineate distinct systemic and local determinants of adipose architecture and provide a structural framework for understanding adipose heterogeneity. Such insights have important implications for fat grafting, regenerative strategies, and interpretation of depot-specific aging or disease-related remodeling. This helps improve our understanding of why adipose depots respond differently to metabolic stress and aging. Furthermore, these data suggest that adipose tissue remodeling cannot be inferred solely from systemic metabolic status but must be interpreted within a depot-specific context.

Purpose: Negative pressure wound therapy (NPWT) improves graft survival and wound vascularity; however, its mechanistic effects on endothelial mechanotransduction remain incompletely defined. We hypothesized that NPWT enhances angiogenic network maturation through activation of the integrin αvβ3–focal adhesion kinase (FAK)–YAP/TAZ signaling axis, promoting endothelial proliferation and vascular stabilization during split-thickness skin graft (STSG) integration.

Methods: A controlled preclinical study was conducted from March 2024 to December 2025 using human endothelial cell–seeded dermal equivalents (n=48 constructs) and a murine STSG model (n=32 grafts). Constructs were exposed to continuous −125 mmHg negative pressure or standard atmospheric pressure for 5 days. Integrin αvβ3 clustering was quantified via immunofluorescence microscopy. FAK phosphorylation (Tyr397) was assessed by Western blot. YAP/TAZ nuclear localization was quantified using confocal imaging. Angiogenic gene expression (VEGF-A, ANGPT2) was measured with qPCR. In vivo, microvascular density, perfusion (laser Doppler), and graft survival were evaluated at 14 and 28 days. A selective FAK inhibitor was used in a subset (n=16 constructs; n=10 grafts) to assess pathway dependence. Only final data are reported.

Results: NPWT significantly increased integrin αvβ3 clustering compared with controls (1.8-fold; p<0.01). Phosphorylated FAK expression increased 2.1-fold under negative pressure (p<0.001). Nuclear YAP/TAZ localization rose by 65% (p<0.01), accompanied by upregulation of VEGF-A (2.3-fold) and ANGPT2 (1.9-fold) (both p<0.01). In vivo, NPWT-treated grafts demonstrated higher microvascular density at 14 days (212±18 vs 146±15 vessels/mm²; p<0.001) and improved perfusion at 28 days (34% increase; p<0.01). Graft survival improved from 78% to 93% (p=0.02). FAK inhibition attenuated YAP/TAZ nuclear translocation and reduced angiogenic gene expression by 45%, abolishing perfusion gains (p<0.05), confirming pathway specificity.

Conclusions: NPWT induces angiogenesis through a defined mechanotransductive cascade involving integrin αvβ3 clustering, FAK activation, and YAP/TAZ nuclear signaling. These findings suggest NPWT functions as a form of surgical mechanotherapy that actively reprograms endothelial signaling rather than solely augmenting perfusion. Targeting this axis may enable pharmacologic or device-based optimization of graft preconditioning and regenerative integration.

References 1. Orgill DP, Bayer LR. Negative pressure wound therapy: past, present and future. Int Wound J. 2013;10 Suppl 1:15-19. 2. Humphrey JD, Dufresne ER, Schwartz MA. Mechanotransduction and extracellular matrix homeostasis. Nat Rev Mol Cell Biol. 2014;15(12):802-812. 3. Chen KD, Li YS, Kim M, et al. Mechanotransduction in response to shear stress. J Biol Chem. 1999;274(26):18393-18400. 4. Zhao B, Tumaneng K, Guan KL. The Hippo pathway in organ size control and tumorigenesis. Nat Cell Biol. 2011;13(8):877-883. 5. Wang KC, Yeh YT, Nguyen P, et al. Flow-dependent YAP/TAZ activities regulate endothelial phenotypes. Nat Commun. 2016;7:10306.

Background: Applicants to integrated plastic surgery residency must create rank lists that determine the next 6+ years of training and life circumstances. While programs increasingly evaluate applicants with structured metrics and consensus processes, applicants commonly rank programs using informal impressions, prestige signaling, and recency bias. Despite the high stakes, there is no standardized, applicant-facing framework for objective, reproducible comparison of programs.

Methods: To describe and propose the Residency Applicant Numeric Key (RANK) tool-a structured, head-to-head scoring framework that enables applicants to compare plastic surgery residency programs across core training and lifestyle domains while allowing individualized priorities. RANK was developed through expert consensus (program leadership and recent graduates) and refined iteratively using feedback from recent applicants obtained through email polling. The tool employs a pairwise (two-program) comparison matrix across 10 domains: case volume, operative autonomy, county hospital experience, community/non-academic exposure, case diversity, faculty and resident culture, global surgery, research volume/mentorship, city/livability, and a flexible "personal" domain (e.g., family proximity, partner considerations, childcare costs, wellness, gut instinct). Applicants assign 1–5 points per domain (5 = most favorable) and may leave domains blank for both programs when insufficiently informed, reducing forced or speculative scoring. Domain scores are summed to generate a total RANK score; the program with the higher total is ranked above the comparator. Applicants repeat pairwise comparisons across all interviewed programs to produce a final ordered rank list.

Results: The RANK framework translates multidimensional impressions into standardized, transparent numeric comparisons and is designed to mitigate recency bias by prompting contemporaneous scoring and repeated head-to-head decisions. The inclusion of an explicit "personal" domain preserves applicant-specific values while keeping the overall process structured. In informal use among consecutive Emory applicant cohorts, users reported improved organization of priorities, clearer differentiation between similar programs, and increased confidence when finalizing rank lists. Anticipated limitations include variability in information available across programs, halo effects from interview-day experiences, and the inherent subjectivity of self-assigned domain scores and weights; these are partially mitigated by recommending scoring within 72 hours of interviews and optional mentor review.

Conclusions: RANK provides a simple, adaptable, applicant-centric tool for structured comparison of plastic surgery residency programs. By decomposing "fit" into observable domains and using repeatable pairwise scoring, RANK offers a practical scaffold that balances analytical rigor with individualized priorities. Future prospective evaluation should assess reliability, usability, and downstream outcomes such as applicant satisfaction and perceived program fit after match.

Purpose: Breast cancer-related lymphedema (BCRL) is a chronic and morbid sequela of breast cancer treatment that significantly impacts patient quality of life, physical function, and psychological well-being. Early identification of high-risk patients can facilitate timely referral for immediate lymphatic reconstruction or early decongestive therapy. Existing BCRL prediction models have limited accuracy and generalizability, and few have utilized machine learning (ML) techniques for predictive modeling. This study aimed to develop and evaluate multiple machine learning (ML) algorithms to predict BCRL using routinely available demographic, clinical, and treatment variables.

Methods: Demographic and clinical data were collected from breast cancer patients undergoing axillary lymph node dissection (ALND) and/or sentinel lymph node biopsy (SLNB) at Memorial Sloan Kettering Cancer Center between 2010 and 2024. Five supervised ML algorithms (Light Gradient Boosting Machine [LightGBM], Extreme Gradient Boosting [XGBoost], Random Forest, AdaBoost, and logistic regression) were trained to predict BCRL. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, F1 score, and Brier score with 95% confidence intervals. Pairwise model comparisons were performed using Wald statistics. Model interpretability was assessed using Shapley Additive Explanations (SHAP) to identify key predictors influencing model outputs.

Results: A total of 20,024 patients were included in the analysis, with a median (IQR) follow-up of 61.7 (38.7–94.2) months. Median (IQR) age was 55 (47–65) years, and median BMI was 25.0 (21.9–29.2) kg/m². Among 3,791 (19%) patients who underwent ALND, 27.8% (1,057) developed BCRL, compared with 4.1% (671 of 16,236) after SLNB. Overall predictive performance was moderate to strong, with AUCs ranging from 0.79 to 0.83. XGBoost demonstrated the highest performance (AUC, 0.83; 95% CI, 0.79–0.83), followed by AdaBoost (AUC, 0.83; 95% CI, 0.79–0.83), while logistic regression achieved an AUC of 0.79 (95% CI, 0.78–0.83). Sensitivity ranged from 0.63 to 0.67, with consistently high specificity across models (0.81–0.84). Models were well calibrated, with Brier scores ranging from 0.17 to 0.21. SHAP analysis identified axillary surgery type, radiation therapy, chemotherapy timing, age, and BMI as key predictors of BCRL.

Conclusion: ML-based prediction models demonstrate reliable performance for identifying patients at elevated risk of BCRL using standard clinical and treatment variables. With further refinement and external validation, these models may support individualized risk stratification and guide early preventive interventions to mitigate lymphedema risk.

Purpose Artificial intelligence (AI) is increasingly used in healthcare for clinical documentation and image generation. Plastic surgery relies on accurate medical illustrations for education and research, but access to trained illustrators is limited by cost and time. Despite rapid advances in generative AI, its ability to produce anatomically precise and educationally useful plastic surgery illustrations remains understudied. This study evaluates whether AI models can generate accurate, clinically relevant medical illustrations for plastic surgery (Davis, 2025). The goal of this study is to provide a framework for integrating AI-generated images into plastic surgery education while maintaining the accuracy and quality of traditional medical illustration.

Methods and Materials Four AI models (ChatGPT Plus, Gemini, Adobe Firefly, and Microsoft Copilot) were used to generate illustrations for six common plastic surgery topics: cleft lip and palate, deep inferior epigastric perforator (DIEP) flap, rhinoplasty, liposuction, breast augmentation, and blepharoplasty. The study included three stages: (1) generation of images using simple prompts and evaluation by medical students, residents, and attending physicians using a standardized rubric; (2) comparison of simple versus advanced prompt image generation using the best-performing model from Stage 1, evaluated with the same rubric; and (3) validation of advanced prompting by additional residents and attendings. Images were scored from 1 (poor) to 5 (excellent) across the following parameters: accuracy, anatomical placement, proportions, color, complexity, educational value, and clinical relevance. Descriptive analysis and paired t-tests compared mean score differences between simple and advanced prompting.

Results Preliminary data from Stage 1 found that ChatGPT outperformed other models, particularly in color accuracy, structural complexity, and overall quality. Mean scores for ChatGPT images ranged from 20.75 to 31.75, with highest scores for rhinoplasty (27.25), saline breast implant placement (30.75), and blepharoplasty (31.75). ChatGPT achieved the highest cumulative score (159.5), narrowly exceeding Microsoft Copilot (153.25) and substantially outperforming Adobe Firefly (92.75) and Gemini (89.25). In Stage 2, advanced prompting with ChatGPT improved mean scores for cleft lip and palate, liposuction, and breast augmentation images; however, differences were not statistically significant (p = 0.33). Cleft lip and palate, DIEP flap, and rhinoplasty required more than 20 additional prompts to achieve ideal image generation standards.

Conclusions AI image generators such as ChatGPT Plus and Microsoft Copilot introduce potential in producing high-quality medical illustrations for simpler anatomical concepts in a faster and more cost-effective manner. However, limitations remain in AI-image generation models, requiring more prompts and containing inaccurate details in more complex anatomical concepts such as DIEP or cleft lip and palate. This suggests the invaluable role trained medical illustrators have in depicting the complexities in plastic surgery anatomy and education. This study provides a framework for future research on integrating AI-generated illustrations into plastic surgery education and research.

Citation 1. Davis P, Napole A, Reddy C, et al. Systematic evaluation of AI-based text-to-image models for generating medical illustrations in neurosurgery: a multi-stage comparative study. Clin Neurol Neurosurg. 2025;257:109039.

Introduction: Adipose-derived stromal cells (ASCs) are widely used in regenerative plastic surgery and require in-vitro expansion to obtain sufficient cell numbers. This culture process, designated by passages (P0, P1, etc.), induces transcriptional and functional changes that may impact therapeutic efficacy. Although prior studies demonstrated progressive loss of stemness and reduced differentiation capacity with later passages, these shifts among early passages remain poorly characterized (1,2). Single-cell RNA (scRNA) sequencing has recently been used to describe ASC heterogeneity at P3, revealing subpopulations with varying degrees of stemness (3). To explore how these subpopulations evolve with subsequent passages, this study uses scRNA sequencing to define shifts from P3 to P4 and clarify their implications for regenerative applications in terms of cell type, cell cycle status and function.

Methods: ASCs were harvested from the stromal vascular fraction and cultured in undifferentiating media DMEM supplemented with FBS and expanded to passages 3 and 4. RNA was extracted using the TRIzol protocol for scRNA sequencing. Clustering and differential expression were analyzed using Loupe Browser (v9.0), and pathway-level alterations between passages were assessed by gene set enrichment analysis (GSEA) in BBrowserX.

Results: A total of 2,354 cells were analyzed at P3 and 2,115 at P4. From P3 to P4, the proportions of ASCs (71.37% to 74.33%) and pericytes (1.36% to 1.70%) increased, and ICAM1⁺ preadipocytes showed a modest rise across their subsets. In contrast, DPP4⁺ multipotent progenitors declined (24.55% to 16.31%), and CD34⁺/CD146⁺ transitional pericytes, present at P3, were no longer detectable at P4. Functionally, adipogenic potential remained absent at both time points, while fibrogenic capacity decreased (84.81% to 66.49%). Chondrogenic (10.25% to 14.43%), osteogenic (4.59% to 18.04%), and myogenic (0.35% to 1.03%) progenitors all increased, suggesting a lineage redistribution rather than diminished stemness. Additionally, early senescent cells increased (55.59% to 69.29%), whereas late senescent cells decreased (43.73% to 30.56%), consistent with culture-associated stress that does not progress to terminal differentiation. GSEA revealed positive enrichment of G1-to-S checkpoint, DNA replication, and pluripotency pathways, alongside negative enrichment of adipogenesis.

Conclusion: Despite prior reports that ASCs lose stemness with continued passaging, this study showed that the P3‑to‑P4 transition produced a lineage shift while maintaining a transcriptionally flexible stromal population. This offers valuable insights into ASC processing strategies applicable for wound healing, and emerging regenerative therapies in plastic surgery.

References: 1. Faghih H, Javeri A, Taha MF. Impact of early subcultures on stemness, migration and angiogenic potential of adipose tissue-derived stem cells and their resistance to in vitro ischemic condition. Cytotechnology. 2017;69(6):885–900. 2. Dao LT, Park EY, Hwang OK, Cha JY, Jun HS. Differentiation potential and profile of nuclear receptor expression during expanded culture of human adipose tissue-derived stem cells reveals PPARgamma as an important regulator of Oct4 expression. Stem Cells Dev. 2014;23(1):24–33. 3. Liu Q, Zhang P, Yuan X, et al. Investigate the stemness of adult adipose-derived stromal cells based on single-cell RNA-sequencing. Cell Biol Int. 2022;46(12):2118–2131.

Background: Autologous fat grafting is widely used for facial reconstruction and contour restoration; however, clinical outcomes remain inconsistent across surgeons, with reports of variable graft survival, contour irregularities, and need for repeated procedures (1). Increasing evidence suggests these outcomes are highly technique-dependent, particularly regarding micro-aliquot placement, tissue plane selection, and avoidance of bolus deposition (2). At present, surgeons typically acquire these skills directly in the operating room because no reproducible simulation platform exists for teaching controlled microcolumn injection.

Objective: To develop and evaluate a low-cost, reproducible silicone-based simulation model that allows trainees and surgeons to practice resistance-guided microcolumn fat injection and visually confirm appropriate micro-aliquot distribution before performing the procedure clinically.

Methods: A transparent silicone polymer model was fabricated to approximate the resistance and compliance of facial soft tissue. The model permits both tactile feedback during cannula passage and direct visualization of injected material. Entry sites were created percutaneously, and injection was performed using 17–20-gauge blunt side-port cannulas attached to 1-mL Luer-lock syringes. Dyed lipid analog material was injected in sequential non-overlapping tracts using micro-aliquots of approximately 0.025–0.05 mL per pass. Distribution patterns were evaluated for column uniformity, separation of tracts, and absence of bolus formation. The model allowed immediate identification of technical errors including tract overlap, excessive volume per pass, and globule deposition.

Results: The simulation reliably demonstrated formation of discrete microcolumns when proper technique was used and clearly revealed technical errors when injection discipline was not maintained. Trainees were able to visually correlate hand pressure, cannula motion, and resistance feedback with resulting distribution patterns. The model reproduced common failure modes observed clinically, including bolus deposition and tract coalescence, allowing immediate corrective feedback. The materials were inexpensive, reproducible, and suitable for repeated practice sessions.

Conclusions: Fat grafting outcomes may be limited by variability in technique acquisition rather than biologic graft failure. This silicone-based simulation model provides a practical platform for teaching controlled micro-aliquot delivery and spatial distribution prior to patient procedures. Simulation-based training may improve procedural reproducibility, shorten the learning curve, and reduce complications associated with early clinical experience in autologous fat grafting.

Citations: 1. Simonacci F, Bertozzi N, Grieco MP, Grignaffini E, Raposio E. Procedure, applications, and outcomes of autologous fat grafting. Ann Med Surg (Lond). 2017;20:49-60. Published 2017 Jun 27. doi:10.1016/j.amsu.2017.06.059 2. Sullivan PK, Palmer SK, Gomez DA. The Evolution of Knifeless Microcolumn Fat Injection in 1362 Patients: Emphasizing Longevity of Results and Optimal Contour. Plast Reconstr Surg Glob Open. 2025;13(9):e7118. Published 2025 Sep 23. doi:10.1097/GOX.0000000000007118

Background: Effective postoperative pain management remains a critical challenge in plastic surgery. Pain remains assessed almost exclusively through intermittent self-report, a method vulnerable to communication barriers, sedation, and reporting bias, contributing to both undertreatment with downstream morbidity and overtreatment with unnecessary opioid exposure. As plastic surgeons emphasize opioid stewardship and enhanced recovery, more precise tools are needed. The objective of this pilot study was to determine whether continuous wearable electrocardiography (ECG) collected during home recovery could be synchronized with patient-reported pain scores to develop and evaluate a machine learning model capable of estimating postoperative pain intensity, establishing foundational evidence for objective, physiology-based pain assessment in plastic surgery and beyond. Methods: This single-arm pilot study was conducted in the Division of Plastic Surgery at Mayo Clinic, Florida, and evaluated continuous physiologic monitoring in postoperative patients using the Mayo Clinic-developed OMNI-H555p wearable device. Adults undergoing outpatient hand surgery, panniculectomy, or hernia repair were enrolled between January 2023 and January 2025. Key exclusions included chronic analgesic use, implanted defibrillator, and documented anxiety or post-traumatic stress disorder. Immediately after surgery, the device was secured over the sternum and worn continuously during home recovery. It recorded ECG over a 72-hour period, along with photoplethysmography (PPG), skin temperature, activity, and environmental signals collected for future multimodal integration. Participants used a study-issued iPhone 13 to report pain intensity via a custom mobile application using a 1–10 Numerical Rating Scale (NRS) four times daily, with additional self-initiated entries permitted. ECG waveform segments were generated and partitioned into training, validation, and testing datasets. All analyses, signal preprocessing, dataset construction, visualization, and machine learning components were implemented within a Python environment. Results: 67 patients completed the study; six were excluded due to withdrawal or data loss. Over the study period, 965 pain questionnaires were recorded (mean 14.4 per subject; SD 4.41). Pain transitions were gradual, and scores were predominantly low (most frequently 1-2) with no documented pain scores of 10. This demonstrated a right-skewed distribution, with a fitted skewed normal function (location = 1.0, scale = 2.14) and overall mean μ = 1.0 (SD = 2.2). A total of 4,904 ECG waveform segments were generated and paired with corresponding NRS values. The training confusion matrix showed complete classification of all samples, indicating 100% training accuracy for the machine learning model. On unseen data, validation and testing accuracies were 58.7% and 59.4%, respectively. Misclassifications occurred predominantly between adjacent pain categories rather than across extreme values, indicating preservation of graded pain signal within ECG-derived features despite class imbalance. Conclusions: This pilot study demonstrates the feasibility of synchronizing continuous wearable ECG data with patient-reported pain scores during home recovery and using these data to train a machine learning model for postoperative pain estimation. The model successfully identified graded physiologic signals associated with pain intensity despite class imbalance and moderate performance on unseen data. These findings provide foundational evidence that wearable autonomic biomarkers may support objective, physiology-informed pain assessment and provide a scalable framework for significant advances in pain management in plastic surgery.

Background: Pulmonary fat embolism syndrome (F-PES) is an uncommon yet potentially life threatening complication of associated with high volume autologous fat grafting. Venous microembolization of adipocytes has traditionally been considered the primary pathogenic mechanism. However, our clinical observations suggest an alternative pathway whereby lymphatic channels and reservoirs from the cisterna chyli become overwhelmed, resulting in the release of lipids into the superior vena cava via lympho venous route. We hypothesize that this lipid trafficking reflects a complex interplay among recipient site interstitial pressure, lymphatic flow dynamics, and systemic triglyceride saturation within the lymphatic and venous circulation. This study also examines a murine model of lipiodol induced F-PES to further elucidate potential lympho venous pathways.

Methods: Twenty seven patients undergoing autologous fat grafting were prospectively enrolled. Peripheral plasma triglyceride concentrations ([TGC]ₚ) were measured preoperatively, intraoperatively, and 24 hours postoperatively. Interstitial hydrostatic pressures were monitored at recipient sites using pressure transducers within subcutaneous and intramuscular compartments. One patient developed clinically and radiographically confirmed F-PES. This patient had central venous catheters placed in the subclavian and femoral veins, permitting sampling from the superior and inferior vena cava. A murine model was created (n = 5) with intraperitoneal Lipiodol injection to assess lipid redistribution via lympho venous pathways via radiographic imaging and histologic analyses of pulmonary tissue and regional lymph nodes.

Results: Across the clinical cohort, peripheral [TGC]p levels remained within a physiological range. In the F-PES case, Central plasma triglyceride ([TGC]c) of the subclavian line was approximately sevenfold higher than corresponding femoral line and peripheral access. The elevated concentrations normalized within 48 hours (p<0.05). Subcutaneous recipient graft sites maintained transient IP elevation below 30 mmHg, resolving within 24 hours, whereas intramuscular regions elevations surpassing 30 mmHg that sustained (p<0.0035). In the murine model, radiographs revealed pulmonary opacities by day five, and histologic sections confirmed intravascular and parenchymal lipid deposition across all specimens.

Conclusion: The combined clinical and experimental evidence indicates that elevated intramuscular pressure following fat grafting may favor passive uptake into lymphatic vessels. This pathway provides a plausible mechanistic addition to direct venous entry in explaining F-PES. The murine findings reinforce the lymphatic-mediated lipid migration and offer the lymphatic conduit as an additional etiology fat embolism.

Background Artificial Intelligence (AI) has emerged as a promising tool to improve research efficiency across different specialties, including plastic surgery. Retrospective data abstraction remains fundamental to research but is labor-intensive and dependent on significant human resources. Despite growing interest in AI-assisted workflows, there are currently limited data evaluating its accuracy compared to traditional human abstraction in academic plastic surgery research. In this retrospective study, we compared AI-assisted and human data abstraction for patient data entered in the National Surgical Quality Improvement Program (NSQIP) Breast Reconstruction pilot program at the University of California – Los Angeles (UCLA).

Methods Human and AI data abstraction accuracy were compared in a randomly selected subset of patients included in the NSQIP breast reconstruction pilot program at UCLA from July 2024 to March 2025. The study only evaluated responses for unstructured variables. Senior surgeon review of completely de-identified patient notes was first performed to create a master key. AI responses were generated and actual NSQIP human abstractor responses were recorded and compared to the senior surgeon review for accuracy. Relative performance was compared overall and stratified by timing, laterality and their interaction. Statistical analysis included McNemar's test, chi-square analysis, prevalence-adjusted bias-adjusted kappa statistics, and mixed-effects modeling accounting for repeated measures within patients.

Results A total of 59 patients (37 unilateral, 22 bilateral) generated 5,054 predefined data points across preoperative, operative, and postoperative time points. Overall performance was high for both reviewers. AI-assisted abstraction demonstrated 99.54% accuracy (95% CI 99.36-99.73%) compared to 99.03% for human abstraction (95% CI 98.76-99.30%) corresponding to 23 and 49 total errors, respectively. Overall agreement between reviewers was 98.65%. McNemar's test demonstrated a statistically significant difference in performance favoring AI (p=0.0016), although the absolute difference in accuracy was small (0.51%). Chi-square analysis demonstrated a significant association between reviewer type and accuracy (p=0.0002). Both reviewers demonstrated lower accuracy for laterality-specific breast variables compared to general patient-level variables, with preoperative variables representing the most challenging. In particular, 9 of the 23 errors made by AI were incorrectly interpreting prior needle biopsy as a history of lumpectomy/excisional biopsy, which was retroactively corrected with 100% accuracy after the completion of the study, demonstrating rapid improvement of the models. AI showed higher accuracy in operative variables involving quantitative measures, whereas human abstraction showed improved performance for selected variables requiring clinical interpretation or historical context.

Conclusion Retrospective data abstraction remains essential to large data research but requires substantial time and personnel resources. In this study, AI-assisted chart abstraction demonstrated superior accuracy compared to human review while simultaneously requiring a fraction of the cost and time. AI abstraction errors were rapidly corrected with iterative model updates. The introduction of AI data abstraction will revolutionize clinical research and large data efforts in plastic surgery.

Purpose: Accurate assessment of the depth and severity of burn injuries is important for surgical decision-making. The use of artificial intelligence (AI) image analysis techniques is being explored as a useful adjunct in the assessment of burn wounds. We conducted a systematic review and meta-analysis to assess the performance of AI-based burn assessment tools.

Methods: PubMed, EMBASE, and Cochrane Library were searched for English-language studies evaluating AI models for burn depth classification or burn area segmentation using clinical or multispectral images. A total of 5,894 records were identified; after screening, 16 studies met inclusion criteria. Eligible studies reported diagnostic performance against clinically meaningful reference standards including expert adjudication, healing outcomes, biopsy, laser Doppler imaging, or clinically validated labels. Random-effects meta-analysis was used to pool sensitivity and specificity. Heterogeneity was quantified using I². Risk of bias and certainty of evidence were evaluated using QUADAS-2 and GRADE frameworks.

Results: 16 studies assessing a total of 44,700 images were included. Of these, 14 studies (37,900 images) evaluated burn depth classification, and 5 studies (7,700 images) also evaluated burn area segmentation (3 studies addressed both tasks). Pooled sensitivity was 0.88 (95% CI, 0.83-0.92) and pooled specificity was 0.94 (95% CI, 0.90-0.97). Most studies reported area under the ROC curve values above 0.90, indicating high discriminative accuracy. There was substantial heterogeneity (I² > 88%), reflecting differences in imaging modalities, AI models, validation methods, and reference standards. The certainty of evidence was rated low to moderate, primarily due to lack of external validation and methodological diversity.

Conclusions: AI-based image analysis is highly accurate for burn wound assessment. Despite strong diagnostic accuracy, more external validation and standardized assessment methods are required to improve confidence in the evidence.