Scientific Abstract Presentations: Research & Technology Session 8 Sunday, October 18 Sun, Oct 18 8:00-9:30 a.m. CDT

Background: Surgical video analysis could support automated workflow understanding, surgical education, and future decision support. However, most general-purpose AI-based computer-vision labeling systems are not tailored to procedure-specific recognition. We evaluated Google Video Intelligence, a general-purpose, cloud-based video annotation API, for detecting procedure-relevant content in nasal basal cell carcinoma (BCC) excision video. We also built an interpretable, rule-based baseline to generate a phase-aware workflow timeline from clinically grounded annotations. Methods: A 1,393.97-second intraoperative video was analyzed, with workflow evaluation restricted to an annotated window from 0 to 1,365 seconds. The video was stored in Google Cloud Storage and analyzed in a Vertex AI environment using Video Intelligence label detection, which returned time-coded label segments (start/end timestamps) with confidence scores. Manual ground-truth annotations were created in parallel with second-level precision. Procedure-relevant instruments and contextual targets were labeled, including scalpel, scissors, forceps, needle driver, suture needle/thread, cautery, specimen, subcutaneous tissue, and gauze-related cues. Label vocabularies from VI and manual annotation were compared. Framewise performance was assessed at 1 Hz on the shared label set using precision, recall, and F1 per label. A deterministic, rule-based phase inference pipeline was then constructed. Manually annotated object/tissue intervals were mapped to six phases: Marking & Setup, Anesthetic Injection, Incision, Excision/Dissection, Specimen Labeling, and Closure/Suturing. Hemostasis was modeled as an overlapping event and quantified within each phase. Results: Manual ground truth captured more procedure-specific content than VI. The manual vocabulary contained 17 unique labels. VI produced 6 unique labels. Overlap was limited to three generic labels: hand, skin, and scar. Most VI outputs reflected broad scene content rather than surgical actions or tools. VI did not reliably return labels for key instruments or targets required for interpreting the workflow in BCC excision. On the shared label set, VI achieved the following framewise results: hand precision/recall/F1 = 0.9079/0.8138/0.8583; skin = 0.9993/1.0000/0.9996; scar = 0.2920/1.0000/0.4520. VI confidence scores were variable (mean 0.36–0.81), with high confidence for skin (0.81) and lower confidence for the remaining labels (0.36–0.47). This suggests that confidence reflects detection strength within the model's label space. It does not imply clinical relevance. The rule-based pipeline produced a clean six-phase timeline across 1,167 seconds. Phase durations were: Marking & Setup 164 s, Anesthetic Injection 69 s, Incision 33 s, Excision/Dissection 360 s, Specimen Labeling 121 s, and Closure/Suturing 420 s. Total hemostasis overlap was 440 seconds. Hemostasis was concentrated in operative work phases, with overlap of 33/33 seconds during Incision, 236/360 seconds during Excision/Dissection, and 110/121 seconds during Specimen Labeling. Overlap was lower during Closure/Suturing (61/420 seconds). Conclusion: This pilot study shows that general-domain video labeling captures broad visual context but misses the procedure-specific signals needed for workflow analysis in BCC excision. In contrast, clinically grounded annotations supported an interpretable six-phase timeline and meaningful event quantification, including overlapping hemostasis. Together, these results highlight the need for surgery-specific label sets and temporally aware models to enable scalable phase recognition across cases.

Background: Successful rhinoplasty outcomes depend on structural grafting and adequate soft-tissue vascular support to enable cartilage integration and long-term stability. Random-pattern flap models are widely used to investigate ischemia, perfusion, and tissue survival mechanisms relevant to reconstructive surgery and have helped define critical ischemia thresholds and healing dynamics.¹–³ Ozone therapy has demonstrated angiogenic, immunomodulatory, and oxidative stress–modulating effects in experimental models.⁴,⁵ This study evaluated the biological impact of ozone therapy on flap survival and healing dynamics as translational evidence supporting soft-tissue conditioning strategies for cartilage graft integration in rhinoplasty. Methods: A prospective experimental study was performed using a dorsal McFarlane random-pattern flap model in 24 male Wistar rats. Animals were divided into six groups (n = 4) according to treatment and sacrifice time point. Outcomes included flap necrosis area, inflammatory infiltrate, fibroblast proliferation, angiogenesis, and expression of hypoxia-inducible factor-1 (HIF-1) and vascular endothelial growth factor (VEGF) assessed by Western blot. Statistical analysis was conducted using the Mann–Whitney U test with significance set at p < 0.05. Results: Ozone therapy significantly reduced flap necrosis compared with controls on postoperative days 2 (p = 0.016), 5 (p = 0.003), 7 (p = 0.003), and 10 (p = 0.003). Increased inflammatory activity was observed in ozone-treated animals at day 10 (p = 0.002), accompanied by higher fibroblast proliferation (p = 0.003). Angiogenesis was significantly enhanced in ozone-treated flaps at day 2 (p = 0.001) and day 10 (p = 0.002). Molecular analysis demonstrated increased VEGF expression in controls at day 2 (p = 0.003), whereas ozone therapy induced early HIF-1 upregulation at day 2 (p = 0.002) followed by decreased expression at day 10 (p = 0.001), suggesting temporally regulated hypoxia-driven reparative modulation. Conclusions: Ozone therapy favorably modulates tissue viability through angiogenesis enhancement, fibroblast activation, and hypoxia-responsive signaling, resulting in improved flap survival. These findings provide mechanistic preclinical evidence supporting ozone-mediated soft-tissue optimization as a potential adjunct approach to improve vascular support and cartilage graft integration in rhinoplasty. Further translational and clinical studies are warranted. References: Zelt RG, Olding M, Kerrigan CL, Daniel RK. Primary and secondary critical ischemia times of myocutaneous flaps. Plast Reconstr Surg. 1986;78(4):498-503. Junior IE, Masson IB, Oshima CTF, Paiotti APR, Liebano RE, Plapler H. Low-level laser irradiation, cyclooxygenase-2 expression and necrosis of random skin flaps in rats. Lasers Med Sci. 2012;27(3):655-660. Gazzalle A, Teixeira LF, Pellizzari AC, et al. Effect of side-stream smoking on random-pattern skin flap survival in rats. Ann Plast Surg. 2014;72(4):463-466. Öksüz M, Yüce S, Koçak ÖF, Canbaz Y, Rağbetli MÇ, Mercantepe T. Effects of ozone pretreatment on viability of random pattern skin flaps in rats. J Plast Surg Hand Surg. 2015;49(5):300-305. Rodríguez ZZ, Guanche D, Alvarez RG, Rosales FH, Alonso Y, Schulz S. Preconditioning with ozone/oxygen mixture induces reversion of oxidative stress indicators and prevents organic damage in rats with fecal peritonitis. Inflamm Res. 2009;58(7):371-375

Purpose: Systematic reviews are valuable to plastic surgery researchers, but abstract screening is time-intensive and variable. Large language models (LLMs) such as ChatGPT may accelerate title/abstract and full-text screening, but optimal prompting is unclear. We evaluated LLM performance for screening in plastic surgery systematic reviews and tested whether PICO statements or inclusion/exclusion (I/E) criteria yield better diagnostic performance. We also compared multiple LLMs to identify the most accurate option.

Methods: A random sample of 50 papers drawn from three completed systematic reviews (human decisions blinded to the LLMs) served as the reference standard. Topics focused on the intersection of plastic surgery and the lymphatic system. We evaluated two screening contexts (abstract and full text). Within each context, prompts supplied either PICO alone, I/E alone, or both. For the combined condition, a paper was excluded only if both the PICO and I/E prompts excluded it. Primary outcomes included confusion matrices and derived metrics (sensitivity, specificity, PPV, NPV, accuracy, and Matthew's correlation coefficient). In a separate comparison on one review, we tested ChatGPT 5, SciSpace Agent, and Claude Sonnet 4.5 using PICO alone, I/E alone, and combined prompting. Agreement with humans and across LLMs was assessed with Cohen's κ and Fleiss' κ, respectively.

Results: For abstract screening, mean sensitivity was 0.61 ± 0.21 (PICO), 0.70 ± 0.15 (I/E), and 0.85 ± 0.14 (combined). For full-text screening, sensitivity was 0.71 ± 0.25 (PICO), 0.56 ± 0.34 (I/E), and 0.90 ± 0.08 (combined). In the model comparison (combined prompting), sensitivities were: ChatGPT 0.86, SciSpace 0.79, and Claude 0.86. Agreement with human screening (Cohen's κ) was 0.71 (ChatGPT), 0.58 (SciSpace), and 0.53 (Claude). Cross-model agreement (Fleiss' κ) was 0.68; including manual screening yielded κ = 0.64. Combining the results of all three LLMs increased sensitivity to 0.93.

Conclusion: LLM-assisted screening can streamline plastic surgery systematic reviews but should be applied cautiously. In our analysis, sensitivity did not meet the Cochrane automated-screening minimum of 0.99. Across abstract and full-text tasks, using both PICO and I/E criteria-and combining their decisions-consistently improved sensitivity over single-prompt approaches, yielding the most favorable sensitivity–specificity balance in our cohort. Among tested models, ChatGPT-5 and Claude Sonnet 4.5 showed similar sensitivities, with SciSpace lower; ChatGPT-5 also achieved higher specificity, PPV, NPV, accuracy, and Matthews correlation coefficient. Combining LLM decisions produced the highest sensitivity, underscoring the value of multi-method screening for improved study capture. All models reliably filtered clearly irrelevant articles but were less adept at identifying all eligible ones, underscoring the need for human oversight. Further work on prompt design and API-level controls (e.g., temperature and other inference parameters) may improve sensitivity. Prospective studies should test these configurations across additional topics and evaluate usability, integration, and reproducibility.

Background: Academic research is crucial to advancing medical knowledge. While there is a trend toward increasing organization and investment in research infrastructure, research in academic centers remains heterogeneous and driven by individual investigators. This is compounded by demanding residents and attending schedules that limit productivity. Preliminary studies suggest that integrating research fellows may improve productivity and academic outcomes. This study aims to evaluate our institutional experience following the implementation of a formal research fellow model and to assess changes in divisional scholarly output and trainee involvement.

Methods: Our previous research model relied on students and residents to complete projects, with students participating either informally or through our institution's yearly summer research program (SRP). In 2020, we implemented a formal research model incorporating 1-3 dedicated research fellows per year while continuing to onboard students through the SRP. In this model, research fellows provided longitudinal project support and facilitated coordination among students, residents, and faculty. We analyzed all published manuscripts and abstracts from January 2015 to January 2025. To account for the delay between implementation of the research fellow-based model and measurable scholarly output, cohorts were defined using a one-year delay: Group 1 (2015-2020) and Group 2 (2021-2025). Data collected included total scholarly output, trainee authorship, and journal impact factors (IF) as a surrogate for publication visibility. Fisher's exact test and chi-squared test were used to compare groups.

Results: From 2015 to 2025, we recorded 94 manuscripts and 284 abstracts. For manuscripts, Group 1 had 26 (28%) while Group 2 had 68 (72%), a 162% increase (p < 0.001). The mean journal IF in Group 1 was 1.5 (SD 0.8) versus 1.8 (SD 0.8) in Group 2 (p = 0.24). In Group 1, students authored 18 (69%) manuscripts, of which 11 (61%) were first authors. In Group 2, students authored 55 (59%) manuscripts, of which 27 (49%) were first authors (p = 0.18 and p = 0.95, respectively). Research fellows authored 46 (49%) manuscripts, of which 18 (39%) were first authors.

For abstracts, Group 1 had 97 (34%) while Group 2 had 187 (66%), a 93% increase (p < 0.001). In Group 1, students authored 48 (49%) abstracts, of which 19 (40%) were first authors. In Group 2, students authored 161 (86%) abstracts, of which 85 (53%) were first authors (p < 0.001 for both, respectively). Research fellows authored 138 (74%) abstracts, of which 45 (35%) were first authors. Overall, individual research fellows averaged 13 (SD 6) manuscripts and 33 abstracts (SD 16).

Conclusions: Implementation of a formal research fellow model was associated with substantial gains in divisional scholarly output and robust trainee participation in academic work. Beyond increasing publication and abstract volume, research fellows may improve project continuity, strengthen research coordination, and expand opportunities for academic engagement and development.

Background: Diabetic wounds suffer from severely impaired angiogenesis, leading to poor tissue regeneration. Impaired wound healing results in tissue necrosis, often limiting reconstructive options. Despite the critical need, no therapeutic modalities that address underlying pathology currently exist to augment surgical wound healing in patients with diabetes. We are pioneering a biologic solution using human adipose-derived multipotent stromal cell conditioned media (ADSC-CM). This cell-derived, yet cell-free approach confers tissue-protective effects similar to stem cell therapy whilst being more pragmatic and having virtually no tumorigenic potential. Our research employs both ex vivo and in vivo methodologies to assess efficacy of ADSC-CM in promoting skin flap survival in a well-established preclinical diabetic mouse model.

Methods:

We characterized ADSC-CM using transmission electron microscopy, nanoparticle tracking analysis, BCA assay, and mass-spectrometry-based proteomics. We performed ex vivo sprouting assays: we incubated 1mm aortic sections from wild type (WT) control and LepRdb/db type 2 diabetic (db/db) mice in matrigel with vascular media or ADSC-CM, and monitored vascular sprouting via microscopy. In vivo, we used a preclinical dorsal peninsular skin flap model, which is separated from fascia with a silicone sheet to study angiogenesis and inosculation from only surrounding skin. We administered subcutaneous saline or ADSC-CM around the flap perimeter and monitored perfusion by laser doppler and gross imaging.

Results:

TEM showed 27.7-160.4 nm particles while NTA revealed 3 peaks of particles with 130, 205, and 313 nm diameter, with 73% of ADSC-CM being 130 nm vesicles. BCA assay revealed a combined concentration of 29.14 mg/mL of protein in ADSC-CM and proteomic analysis showed enrichment for cell motility and angiogenesis-associated molecules. Aortic rings from db/db mice in ADSC-CM had significantly increased radial sprouting by 10 days compared to db/db rings in vascular media (1.08 mm2 ± 0.13 mm2 vs 0.17 mm2 ± 0.47 mm2, respectively (mean ± SEM), p<0.05, n=6), indicating that ADSC-CM promotes angiogenesis from existing endothelial cells despite diabetes-induced impairment. Laser doppler imaging showed ADSC-CM-treated db/db mice had a 71% increase in flap perfusion 8 days after flap creation, while saline-treated WT mice had 41% less perfusion over the same interval. Gross imaging showed significantly improved area of flap survival in ADSC-CM-treated flaps compared to saline-treated db/db counterparts over 14 days (78.2 ± 8.5% vs 42.5 ± 8.1%, respectively, mean ± SEM, p<0.05, n=8), likely due to successful inosculation and reperfusion from adjacent intact skin.

Conclusion:

Skin flaps in diabetic mice receiving local ADSC-CM therapy showed improved viability and perfusion. These findings, supported by ex vivo sprouting assays, suggest that ADSC-CM promotes angiogenesis and enhances flap survival. Our outcomes demonstrate that ADSC-CM shows promise as an adjunct treatment in patients with diabetes where flap viability limits reconstructive options.

Purpose: Complex wound management requires rapid synthesis of evolving evidence across heterogeneous pathologies, yet decision-making often relies on fragmented literature review or non-transparent AI tools. Existing large language models lack source attribution and may generate unsupported recommendations. We developed and evaluated a multimodal Retrieval-Augmented Generation (RAG) system designed to provide real-time, literature-grounded clinical decision support for complex wound care.

Methods: A wound-specific knowledge base of 5,796 open-access full-text PubMed publications was curated and indexed using semantic embeddings. The RAG architecture integrates targeted document retrieval with large language model reasoning, constraining outputs to retrieved literature and providing citation transparency. The system supports multimodal image input to allow context-aware wound assessment. Performance was prospectively evaluated across 10 clinically representative wound scenarios spanning acute, chronic, and emergent conditions, including negative pressure wound therapy selection, pressure ulcer debridement, diabetic foot ulcer differentiation, skin graft bed preparation, necrotizing soft tissue infection management, hyperbaric oxygen therapy indications, and acute burn care with compartment syndrome. Primary evaluation metrics included semantic similarity to reference literature, contextual accuracy (G-score), citation alignment, and qualitative assessment of recommendation completeness, including contraindications and timing considerations.

Results: Across all scenarios, the system consistently retrieved 5–9 relevant publications per query and generated literature-grounded recommendations with high semantic similarity (range 0.86–0.96; mean 0.93) and contextual accuracy (range 0.80–1.00; mean 0.95). Outputs incorporated key management domains including operative timing, debridement strategy, antibiotic selection, adjunctive therapies, and contraindications. Citation alignment was maintained across all responses without unsupported claims. Performance remained stable across emergency and elective contexts, and multimodal image input enabled contextual refinement of recommendations in visually dependent cases.

Conclusion: A wound-specific, multimodal RAG-based AI system can generate transparent, evidence-grounded recommendations across diverse wound care scenarios. By constraining outputs to retrieved literature and preserving citation traceability, this approach addresses key safety limitations of generic AI models. These findings support the feasibility of literature-anchored AI decision support to augment clinical reasoning in complex wound management and warrant prospective clinical validation.

Background: Large language models (LLMs) are increasingly explored for clinical decision support and patient education [1]. In reconstructive plastic surgery, accurate postoperative guidance is critical for wound care, flap monitoring, drain management, scar optimization, and recognition of complications [2]. However, integrating specialty-specific knowledge into LLMs while maintaining safety and reliability remains challenging. Fine-tuning (FT), retrieval-augmented generation (RAG), and hybrid FT+RAG approaches represent distinct strategies for domain adaptation, yet their comparative performance in reconstructive postoperative care has not been systematically evaluated [3-5].

Objectives: We aimed to compare the performance, reliability, and safety characteristics of baseline, fine-tuned, retrieval-augmented, and hybrid FT+RAG LLM configurations for postoperative decision support.

Methods: We conducted a controlled comparative evaluation of four LLM configurations using Gemini 2.5 Flash: (1) baseline model, (2) fine-tuned model, (3) RAG-based model grounded in a curated reconstructive postoperative knowledge base, and (4) hybrid FT+RAG model. A total of 600 expert-authored reconstructive postoperative question–answer pairs were used for model adaptation and validation, with 150 held-out queries reserved for final evaluation. Queries included routine reconstructive care questions (e.g., wound care, drain management, activity restrictions), emergency scenarios (e.g., flap compromise, infection), and out-of-scope items. Three blinded clinical reviewers assessed medical accuracy, completeness, and relevance. Automated metrics evaluated readability, faithfulness, and hallucination propensity.

Results: All knowledge-enhanced models significantly outperformed baseline in overall accuracy (68.0% vs FT 92.7%, RAG 91.3%, FT+RAG 97.3%; p<0.001). The hybrid FT+RAG configuration achieved the strongest classification performance (precision 100%, recall 96.7%, F1 98.3%). FT and RAG alone produced comparable improvements in clinical accuracy and factual alignment. Completeness scores were higher in all knowledge-enhanced models, whereas relevance remained consistently high without significant differences across configurations. Enhanced models generated shorter, more focused responses; however, readability exceeded recommended patient education thresholds. Faithfulness improved and hallucination rates decreased significantly in all knowledge-augmented models compared with baseline (p<0.001).

Conclusions: In reconstructive plastic surgery postoperative care, integrating domain-specific knowledge into LLMs substantially improves clinical accuracy and safety relative to baseline models. Hybrid FT+RAG architectures provide the most favorable balance of performance and grounding, although gains over single-strategy approaches were modest. These findings support the potential role of knowledge-augmented LLMs in reconstructive postoperative education, while emphasizing the need for readability optimization, transparency, and sustained clinician oversight prior to clinical deployment.

References: 1. Wang, M., H. Ma, and M. Piao, Effectiveness of large language models in preoperative and discharge education: a systematic review based on an evaluation framework. npj Digital Medicine, 2026. 2. Knoedler, S., et al., Postoperative free flap monitoring in reconstructive surgery-man or machine? Frontiers in Surgery, 2023. 10: p. 1130566. 3. Yang, W., et al., A comprehensive survey on integrating large language models with knowledge-based methods. Knowledge-Based Systems, 2025: p. 113503. 4. Lewis, P., et al., Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems, 2020. 33: p. 9459-9474. 5. Soudani, H., E. Kanoulas, and F. Hasibi. Fine tuning vs. retrieval augmented generation for less popular knowledge. in Proceedings of the 2024 Annual International ACM SIGIR Conference on Research and Development in Information Retrieval in the Asia Pacific Region. 2024.

Background: Aesthetic surgery tourism has seen a significant increase in recent years. However, patients undergoing long-haul flights after surgery may be at higher risk for complications due to physiological changes associated with air travel. This study examines the impact of commercial flight conditions on postoperative patients traveling after aesthetic procedures. Methods: This prospective study evaluated physiological variables in postoperative patients undergoing aesthetic surgery tourism during flights exceeding four hours: oxygen saturation, heart rate, and popliteal vein diameter were monitored throughout the flight. The findings were compared to a control group of non-surgical travelers. The study also analyzed the potential contribution of these physiological changes to postoperative complications. Results: All postoperative patients experienced oxygen desaturation, with an average drop of 4% to 7% in hemoglobin saturation within the first four hours of flight. Postoperative patients also exhibited significantly lower baseline hemoglobin levels compared to the presurgical levels. Heart rate increased by 15–20 bpm in all participants, and popliteal vein diameter progressively expanded throughout the flight due to hypoxia-induced vasodilation and prolonged sitting. These physiological alterations align with risk factors for thrombosis, flap necrosis, infection, and postoperative bleeding. Conclusion: Patients undergoing aesthetic surgery tourism experience decreased oxygen saturation, increased heart rate, and an enlarged popliteal vein diameter due to the body's intrinsic responses to hypoxia. These physiological changes, combined with their postoperative state, may increase the risk of complications such as thrombosis, necrosis, infection, hematomas, and bleeding. Postoperative care protocols and pre-flight counseling are essential to ensure patient safety and mitigate these risks.

Background: Microsurgical reconstruction is technically demanding and vulnerable to errors, such as accidentally catching the posterior wall during vessel anastomosis ("back-walling"). Inadvertent back-wall stitching increases thrombosis risk and contributes to free flap failure rates observed as high as 10%. We developed the Equidistant Anastomosis Suture Instrument (EASI), a novel microsurgery instrument that enhances anastomosis quality, decreases duration of vessel anastomosis procedures, and prevents inadvertent back-wall stitching to improve flap outcomes. Utilizing an ex-vivo vessel model, we evaluated whether using EASI as an intraluminal guide improves efficiency and technical performance, compared to traditional anastomosis suturing technique, in trainees and senior plastic surgeons.

Methods/Materials: An ex-vivo chicken thigh vessel anastomosis model was used to compare traditional vessel anastomosis technique and EASI-assisted vessel anastomosis. Study participants included trainees comprising of medical students and residents, as well as senior attending physicians consisting of board-certified plastic surgeons. Outcomes were: (i) anastomosis duration (mm:ss), (ii) self-rated confidence in no back-walling immediately post-procedure (1–10; 1 = "definitely back-walled", 10 = "definitely did not back-wall"), and (iii) external appearance and quality of vessel anastomosis score (1–10) determined by a microsurgery fellowship-trained plastic surgeon.

Results: Thirty-two anastomoses (traditional n = 17, EASI n = 15) were analyzed for duration and confidence in no back-walling scores, while external appearance and quality of vessel anastomosis scores were analyzed for 17 anastomoses (traditional n = 11, EASI n = 6). Across all participants, EASI reduced mean anastomosis completion time by 29% compared with traditional anastomosis technique (20:10 ± 8:01 versus 28:13 ± 16:26, respectively). Among senior plastic surgeons, mean completion time particularly favored EASI with a 34% reduction in anastomosis duration over traditional technique (9:27 ± 3:14 versus 14:22 ± 2:25), while trainees demonstrated a 23% reduction in anastomosis duration using EASI versus traditional anastomosis (24:04 ± 4:54 versus 31:11 ± 16:39, respectively). Confidence in no back-walling scores were also substantially higher in all participants using EASI (9.67 ± 0.62) versus traditional technique (6.06 ± 1.20), with trainees demonstrating significantly greater confidence in no back-walling when using EASI compared to traditional anastomosis (9.55 ± 0.69 versus 5.86 ± 1.23; participant-adjusted analyses p<0.001). One anastomosis was confirmed to have back-walling in the traditional anastomosis group. External appearance and quality of vessel anastomosis scores were higher for all participants using EASI (9.17 ± 1.33) versus traditional anastomosis (5.09 ± 3.08; nonparametric comparison p=0.008). Particularly for trainees, the average external appearance and quality scores were more than 2-times greater when using EASI, compared to traditional anastomosis (3.63 ± 2.13 versus 7.50 ±0.71, respectively).

Conclusions: In an ex-vivo chicken thigh model, EASI was associated with faster anastomoses, markedly higher confidence in avoiding back-walling, and superior external appearance and quality of anastomosis scores versus traditional anastomosis suturing technique. This preliminary data supports EASI as a potential industry-changing device for advancing technical performance, procedural standardization in microsurgery, and reducing error-prone variability, ultimately improving patient outcomes. Future in-vivo validation studies will further define EASI's performance and ascertain its identity as a high-impact clinical tool and powerful adjunct in microsurgery.

BACKGROUND: Plastic surgeons utilize creativity and artistry to manipulate tissues to restore the human form. Pedagogical models that integrate arts into surgical curricula like museum-based observation and hands-on workshops have demonstrated benefits in cultivating observational skills, visuospatial reasoning, and aesthetic awareness (1–4). However, the extent to which visual and digital arts education have been formally incorporated into plastic surgery training remains unclear. This study aimed to characterize existing arts-based curricula in plastic surgery residency programs while also assessing program directors' attitudes towards their educational value and implementation hurdles.

METHODS: A cross-sectional 33-item needs assessment was electronically distributed to program leadership of the 89 integrated plastic surgery residency programs in the United States to evaluate: (a) current implementation and perceptions of arts-based curricula in plastic surgery residency, (b) importance of visual and digital arts competencies in a standardized arts curriculum, and (c) significance of barriers to implementation. Descriptive statistics were used to analyze responses, and consensus on 5-point Likert scale questions across all domains was defined as ≥75% agreement.

RESULTS: Survey responses were received from 20 program directors (22.5% response rate), showing that 70.0% have already incorporated visual and digital arts in their program curricula, most commonly through didactics on aesthetic analysis (57.1%) or photography (50.0%). Despite this, 30% disagreed that their program's graduating residents were competent in digital aesthetic arts like the use of computer visualization and morphing tools on patient photographs. Nearly half of respondents (45.0%) agreed on the importance of arts training and expressed interest in incorporating such curricula, with the remaining half reporting neutral opinions. Program directors reached consensus agreement that it is important for a standardized arts curriculum to teach observational skills, visuospatial awareness, clinical photography techniques, visual teaching, visual-motor integration, aesthetic judgement, and dexterity and fine motor skills. None of the barriers to implementation rose to consensus agreement, but respondents most frequently cited lack of curricular time, lack of established models, and insufficient faculty expertise.

CONCLUSION: Visual arts and digital aesthetics education have been adopted by many integrated plastic surgery programs, but curricular implementation remains variable with specific concerns regarding digital media competency among residents. Program directors achieved consensus regarding several high priority curricular components but cited curricular time and resource constraints as barriers. This needs assessment supports the need for development of a standardized, evidence-based arts curriculum to promote consistent training in visuospatial, aesthetic, and digital competencies within plastic surgery residency.

REFERENCES: 1. Cohen SM, DiGiovanni-Evans B, Ganske IM, Katz JT, Kent TS. Sewing the SEAMs: Surgical Education in the Art Museum. J Surg Educ. 2025;82(3):103401. doi:10.1016/j.jsurg.2024.103401 2. Cohen SM, Dai A, Katz JT, Ganske IM. Art in Surgery: A Review of Art-based Medical Humanities Curricula in Surgical Residency. J Surg Educ. 2023;80(3):393-406. doi:10.1016/j.jsurg.2022.10.008 3. Güneron E, Kivrak N, Koyuncu S, Tuncer S, Uysal A. Aesthetic Surgery Training: The Role of Art Education. Aesthet Surg J. 2005;25(1):84-86. 4. Abdulwadood I, Anderson M, Kough K, Akridge A, Noland SS. Surgical Education: Integrating Visual Thinking Strategies into Plastic Surgery Training. Plast Reconstr Surg. 2024;154(4):848e-850e. doi:10.1097/PRS.0000000000011461

Background Androgenic alopecia (AGA) is the most common cause of hair loss in both men and women and is characterized by progressive follicular miniaturization and perifollicular inflammation1. Current first line medical therapies, including minoxidil and finasteride, provide limited results. Regenerative treatments such as stromal vascular fraction (SVF) and platelet rich plasma (PRP) have demonstrated promising effects2-3. However, comparative, controlled clinical data remain limited. This study aims to evaluate and compare the efficacy of SVF, PRP and microneedling in the treatment of AGA. Methods This is a prospective, interventional, randomized, single blinded, controlled trial including 75 healthy male patients aged 18–55 years with androgenic alopecia (Norwood–Hamilton grades III–IV). Patients were randomized into three equal groups: SVF scalp injection, PRP scalp injection and Microneedling with saline injection. SVF was obtained from autologous abdominal lipoaspirate using a mechanical processing system (Lipocube), while PRP was prepared from autologous venous blood (Cellenis). All treatments were administered via standardized subcutaneous scalp injections. Patients were evaluated at treatment, 3 and 6 months. Primary outcomes include objective hair density, follicle count, and hair shaft diameter measured using dermoscopy. Secondary outcomes include standardized clinical photography, investigator assessment using the Norwood–Hamilton scale, and patient reported satisfaction using validated questionnaires (HairQ and Subject Hair Satisfaction Questionnaire). Results Significant miniaturization reversal was observed through increased hair shaft diameters and enhanced melanin concentration, effectively improving scalp masking. Furthermore, post-treatment normalization of the perifollicular environment and reduced erythema indicate a stabilized substrate for sustained terminal hair growth in all three groups. All three groups demonstrated a reduction in hair shedding during the follow up period. Objective analysis revealed an improvement in hair density and follicular units density in the treatment groups. Hair density increased 14% in the microneedling group, 16% in the PRP group and 17% in the SVF group, compared with baseline all statistically significant (p<0.001). Comparison of density improvement between 3 groups was not significant (p=0.3). Hair follicle diameter increased following treatment in all therapy groups. Patient reported satisfaction scores were significantly higher in the PRP and SVF groups compared with the saline control group, demonstrating a statistically significant difference in subjective outcomes between PRP, SVF and microneedling cohorts.

Conclusions Microneedling, PRP, and SVF all significantly reverse follicular miniaturization and enhance hair density. SVF yielded the greatest objective improvements, followed by PRP and microneedling, respectively. Furthermore, patient satisfaction was significantly higher in the SVF and PRP cohorts compared to microneedling. SVF represents the most effective regenerative approach for androgenic alopecia.

References 1. Jang WS, Son IP, Yeo IK, et al. The annual changes of clinical manifestation of androgenetic alopecia clinic in Korean males and females: an outpatient-based study. Ann Dermatol. 2013. 2. Bourin P, Bunnell BA, Casteilla L, et al. Stromal cells from the adipose tissue–derived stromal vascular fraction and culture-expanded adipose tissue–derived stromal/stem cells. Cytotherapy. 2013. 3. Epstein GK, Epstein JS. Mesenchymal stem cells and stromal vascular fraction for hair loss: current status. Facial Plast Surg Clin North Am. 2018.

Purpose: Rhinoplasty outcome assessment relies predominantly on subjective surgeon judgment and qualitative photographic comparison, which suffer from poor inter-rater reliability and lack standardized benchmarks. Artificial intelligence (AI) offers the potential to standardize this process through automated morphometric quantification. This systematic review and meta-analysis evaluate the accuracy, reproducibility, and agreement with expert assessment of AI models applied to objective morphometric outcome measurement in rhinoplasty.

Methods: We searched PubMed/MEDLINE, IEEE Xplore, Scopus, and the Cochrane Library from inception through January 2025. Included studies applied AI or machine learning to 2D photographs or 3D surface images for quantitative measurement of post-rhinoplasty morphometric outcomes, including anthropometric distances and angles, symmetry indices, and proportionality ratios. Twelve studies (2018-2025; n=23-3,000 patients per study) met inclusion criteria. Risk of bias was assessed using modified QUADAS-AI. Random-effects meta-analysis was performed for continuous morphometric error metrics; classification accuracy was summarized descriptively due to task heterogeneity.

Results: AI models achieved a descriptive pooled classification accuracy of 76.7% (95% CI: 65.2-88.2%; range 52.5-89%), with 3D-based models outperforming 2D systems (84.5% vs. 69.2%). Mean landmark localization error was 1.38 mm (95% CI: 0.94-1.82 mm), approaching sub-millimeter precision at optimal landmarks (pronasale: 1.18 ± 1.10 mm). Morphometric measurement reproducibility was high (ICCs: 0.72-0.96), exceeding typical human inter-rater reliability (0.60-0.85). AI-surgeon agreement for simulation-based outcome evaluation was 68.4% (95% CI: 64.9-71.7%), significantly lower than inter-surgeon agreement of 77.3% (95% CI: 74.2-80.3%; p<0.0001). Sensitivity analysis excluding high risk-of-bias studies (n=3) yielded 78.9% accuracy (95% CI: 70.1-87.7%) with reduced heterogeneity (I²=71% vs. 82%). GRADE certainty was low.

Conclusion: AI demonstrates high reproducibility and clinically approaching accuracy for quantifying objective morphometric outcomes in rhinoplasty, particularly with 3D imaging inputs. However, AI does not yet replicate integrative surgeon judgment, and the evidence is constrained by methodological heterogeneity, single-center datasets, and the absence of prospective validation. Clinical translation requires consensus on standardized morphometric endpoint definitions, multicenter prospective trials, and integration of objective AI-derived metrics with validated patient-reported outcomes.

Purpose: Tissue expanders are commonly used in two-stage implant-based breast reconstruction to gradually create space for a permanent implant by stretching the overlying soft tissue. To optimize outcomes, surgeons frequently place acellular dermal matrix (ADM) at the time of expander placement to provide inferolateral support, offload tension from mastectomy flaps, and reduce capsular fibrosis. Despite its widespread use, the mechanisms by which ADM modulates capsule formation in the setting of mechanical tissue expansion remain poorly defined. Furthermore, in clinical practice, expanders are initially filled with either air or saline, but the impact of the inflation medium on capsule formation remains unclear. We developed a clinically relevant murine tissue expander model incorporating sustained mechanical stimulation to compare air versus saline fill and to evaluate the mechanoprotective effect of ADM. Methods: Miniature smooth tissue expanders were combined with our mechanically stimulating implant (MSI) model to reliably induce clinically relevant, severe foreign body response (FBR)(1,2). Expanders were then filled with either air or saline, with parallel groups receiving ADM coverage, and implanted into the dorsal subcutaneous space of mice (n = 5/group). Capsule tissue was harvested on day 28. We quantified capsule thickness, collagen deposition, α-smooth muscle actin (α-SMA)–positive myofibroblasts, and immune cell infiltration, including CD45⁺ immune cells and F4/80⁺ macrophages. Results: Mechanical stimulation produced a robust and reproducible fibrotic capsule consistent with clinically observed fibrosis. Air and saline inflation yielded similar outcomes, with no significant differences in capsule thickness (p = 0.71-0.84), collagen content (p > 0.99), α-SMA⁺ cells (p > 0.99), CD45⁺ cell infiltration (p = 0.24-0.99), or F4/80⁺ macrophage density (p > 0.99).　In both air and saline filled devices, ADM coverage significantly reduced capsule thickness (p < 0.005), collagen deposition (p < 0.001), α-SMA⁺ myofibroblasts (p < 0.05), CD45⁺ immune infiltrates (p < 0.001), and F4/80⁺ macrophages (p < 0.005) compared to expanders without ADM coverage. Discussion: We present the first murine tissue expander model that reliably induces mechanically driven capsule formation analogous to clinical FBR. Using this model, we found that air and saline inflation produce comparable levels of capsule formation. Notably, we found that ADM confers clear mechanoprotective effects by decreasing capsule fibrosis and inflammation. Although ADM has been clinically associated with improved outcomes, its biological mechanism has remained uncertain. This model provides direct experimental evidence that ADM attenuates the fibroinflammatory response generated during expansion. Our findings offer a mechanistic justification for incorporating ADM coverage at the time of tissue expander placement.

1. Padmanabhan J, Chen K, Sivaraj D, et al. Allometrically scaling tissue forces drive pathological foreign-body responses to implants via Rac2-activated myeloid cells. Nat Biomed Eng. 2023;7(11):1419-1436.
2. Sivaraj D, Padmanabhan J, Chen K, et al. IQGAP1-mediated mechanical signaling promotes the foreign body response to biomedical implants. FASEB J. 2022;36(2):e22007.

Background: Patient and caregiver education remains a persistent challenge in cleft lip and palate (CLP) care, where complex, staged treatment pathways intersect with wide variability in caregiver health literacy. Our prior work evaluating large language models (LLMs) for cleft lip counseling demonstrated that general-purpose models could generate understandable responses but also revealed important limitations, including variable accuracy, lack of sourcing, and limited clinical specificity when used without constraints. These findings motivated the development of a more structured approach that preserves the accessibility of conversational AI while improving reliability. Here, we describe the development and early feasibility evaluation of CLARA (Cleft Lip and Palate Assistant for Resources and Answers), a retrieval-augmented AI platform designed to support consistent caregiver education in cleft care.

Methods: A retrieval-augmented generation (RAG) framework was developed using a curated knowledge repository composed of standardized cleft lip and palate educational materials and frequently encountered caregiver questions. Relevant content was retrieved for each user query and provided to a large language model to support response generation grounded in these curated materials. A web-based interface enabled structured internal testing. Iterative review of representative responses informed ongoing refinement of both the knowledge base and the response generation process.

Results: During feasibility testing, the retrieval-augmented system produced responses that were more consistent and clinically aligned compared with unrestricted LLM output. Responses demonstrated improved uniformity of messaging, reduced speculative language, and more appropriate terminology for caregiver education. No unsupported clinical recommendations were identified.

Conclusions: A retrieval-augmented conversational AI framework may help support more consistent caregiver education in cleft care while maintaining the accessibility of interactive language models. Building on prior evaluations of unguided LLMs, this approach represents a more controlled and clinically aligned model that may be adaptable to other clinical domains. Future work will focus on expanding the knowledge repository and conducting caregiver-facing evaluations of usability and comprehension.