From Static Atlases to Intelligent Learning Systems: A Scoping Review of Imaging-Based Tools and Emerging Artificial Intelligence Applications for Facial Aesthetic Education and Training

Abstract

Aesthetic surgery training requires anatomical knowledge, psychomotor skill, aesthetic judgment, and patient-centered decision making. Traditional training pathways face constraints as demand for aesthetic procedures increases and unregulated practice exposes patients to preventable complications. Multimodal imaging, simulation, and artificial intelligence (AI)-enabled tools may support safer, scalable training, but educational validation remains heterogeneous. This scoping review maps imaging-based tools and emerging AI applications relevant to facial aesthetic surgery education and training, including direct facial aesthetic studies and transferable technical advancements from the general reconstructive subspecialties. PubMed, MEDLINE, Embase, and the Cochrane Library were searched for English-language primary studies published from January 1, 2016, to February 28, 2026. Studies were screened by 2 reviewers and mapped by imaging scale, educational function, learner outcome, validation status, and study classification. Of 106 full-text reports assessed, 16 studies were included: 15 were in plastic surgery or facial anatomy contexts and 1 was in dermatology. Current evidence supports imaging-based visualization and simulation as promising adjuncts, but objective educational validation remains limited. Overall, this review provides a synthesis of aesthetic surgery training tools and platforms that support the training process. The continued development of high-quality imaging datasets and validated AI-assisted tools may provide an important foundation for future anatomy and aesthetic surgery education.

Level of Evidence: 5 (Diagnostic)

Surgery differs fundamentally from most other medical specialties. It requires not only cognitive understanding but also complex psychomotor skills and multisensory integration. Mastery of surgical techniques relies on the ability to interpret tactile and visual feedback, including tissue resistance, texture, pulsation, and color, which are difficult to convey through traditional text-based or didactic teaching.¹ Consequently, surgical training remains heavily dependent on hands-on experience, limiting learning opportunities in the absence of simulation platforms or supervised operative exposure.²

In plastic surgery, particularly in face and neck procedures, this challenge is further amplified. Interventions must address symmetry and proportion, restore or modify muscle position and function, and take into consideration the psychosocial impact of morphological and aesthetic changes.³ Such complexity requires a more precise understanding of anatomy and its dynamic relationship to appearance.³ To achieve safe and effective results, the training process of a plastic surgeon is hence particularly time- and resource-intensive, posing challenges for efficient training in the context of rapidly increasing patient demand. As a result, educators and trainees face a persistent tension between training efficiency and patient safety.

To address these challenges, institutions are exploring innovative educational approaches that integrate digital technologies into surgical training. Multimodal imaging has emerged as a promising tool set for advancing facial anatomical research, surgical education, and perioperative planning.^4-6 Techniques such as 3-dimensional (3D) photogrammetry, computed tomography (CT), MRI, and high-resolution ultrasound enable detailed visualization of facial structures across multiple anatomical scales. Recent studies have also reported automated recognition systems and imaging-based decision-support tools that enhance clinical guidance.^7,8

Beyond imaging technologies, the rapid development of artificial intelligence (AI)-enabled learning systems introduces new opportunities for surgical education.^9,10 Compared with traditional digital resources, these systems may support high-frequency, personalized learning by integrating image interpretation, anatomical questioning, procedural reasoning, and feedback mechanisms.^11,12 Such platforms have the potential to help trainees translate multimodal anatomical representations into actionable operative knowledge. Furthermore, high-quality data derived from expert practice may serve as a foundation for the development of next-generation intelligent systems, potentially extending toward agent-assisted or semi-autonomous surgical applications.

However, most of these technologies are primarily designed to assist experienced clinicians rather than to support structured educational pathways for trainees. As a result, the growing body of multimodal anatomical knowledge remains insufficiently integrated into surgical education frameworks.

This scoping review therefore aims to map and critically evaluate imaging-based tools and emerging AI applications relevant to facial aesthetic surgery education and training, with particular attention to their educational functions, learner populations, reported outcomes, validation status, and remaining gaps in evidence. The review centers on facial aesthetic surgery education and includes nonsurgical facial aesthetic procedures, including procedures from craniofacial, maxillofacial, and selected facial imaging or simulation studies when they address directly shared anatomy or procedural guidance based on the defined transferability criteria.

METHODS

Databases and Search Engines Utilized

This review was designed as a scoping review, using a PRISMA-ScR-informed approach to map available technologies, educational applications, and evidence gaps.¹³ Electronic databases including PubMed, MEDLINE, Embase, and the Cochrane Library were searched for English-language articles published from January 1, 2016, to February 28, 2026; the final search was performed on February 28, 2026. Database searches were conducted by (M.Z. and Y.M.). The time frame was selected to capture major technological advances post-2010 and PubMed-indexed publication trends indicating increased uptake of simulation, virtual or augmented reality (AR), and robotics in surgical training starting from the mid-2010s. The database strategy was anchored to facial aesthetic and plastic surgery, facial anatomy, surgical education, and imaging. Artificial intelligence–related studies were eligible when linked to imaging, simulation, or educational use. Full database search strings are provided in Supplemental Table 1.

Eligibility Criteria

Studies were eligible for inclusion if they were quantitative or qualitative investigations that incorporated at least one imaging-based component. For this review, imaging modalities were categorized by the anatomical scale they primarily depict: (1) macro-level imaging, capturing whole-face or region-scale 3D anatomy, and deeper skeletal and soft-tissue relationships (CT, MRI, 3D photogrammetry, cadaveric volumetric reconstruction), (2) meso-level imaging, resolving localized tissue layers and structural details (ultrasound, optical coherence tomography angiography, and vascular mapping techniques), and (3) micro-level imaging, depicting micrometer, and submicrometer anatomical structure (micro-CT, optical coherence tomography, 2-photon microscopy, dermatoscopy, histologic, or plastination-based structural analyses).

Studies were required to address education and training in at least one of the following nonsurgical or surgical facial aesthetic domains: facial aesthetics, craniofacial, dermatological, or maxillofacial surgery. For this review, transferability of relevant imaging or AI tools was screened during full-text assessment by inspecting whether the technology addressed at least one training function directly shared with facial aesthetic practice: (1) layered facial anatomy visualization, (2) image-guided localization or planning of facial procedures, (3) simulation or rehearsal of facial procedures with similar spatial or psychomotor demands, or (4) image-based decision support or feedback applicable to trainee interpretation.

To maintain focus, studies from relevant subspecialties were included only when their technology could be translated to facial aesthetic procedural training through anatomy visualization, image-guided procedural learning, simulation, decision-support training, or feedback. For educational studies, eligibility required a qualitative or quantitative education- or training-related outcome, including knowledge gain, anatomical understanding, psychomotor skill acquisition, procedural accuracy, learner confidence, satisfaction, feedback, or aspects of professional identity formation. Studies describing multimodal image fusion, deep learning applications, or AI-assisted modeling were considered eligible only when aligned with the above criteria.

Exclusion criteria were non-English articles, conference abstracts, editorials, commentaries, and short communications without primary data.

Selection of Reports

Two reviewers (M.Z. and Y.M.) independently screened titles and abstracts for relevance. Full-text articles were subsequently assessed according to predefined inclusion and exclusion criteria. Discrepancies were resolved through discussion and consensus; any persistent disagreement was planned for adjudication by a third senior reviewer (N.Y.), although no unresolved discrepancies remained after consensus.

Data Extraction and Study Classification

Two reviewers (M.Z. and Y.M.) extracted study design, specialty, learner group, imaging modality, AI component, educational function, outcome type, validation approach, and main limitations. Studies were grouped into 3 study classifications: educational intervention studies, training platform studies, and educationally relevant technology studies. These classifications were developed deductively from the review question and refined inductively after full-text review. The classification was used to distinguish the role each study played in this scoping review. Assignment was based on the primary purpose of each study: direct learner assessment, platform development for training or rehearsal, or technology with clear but indirect educational value.

Methodological Appraisal

Because the included literature contained randomized educational trials, observational educational studies, simulation platform reports, and technical feasibility studies, a single formal risk-of-bias instrument was not applicable across all studies. Instead, methodological rigor was assessed descriptively by study design, learner population, outcome level, validation status, and whether outcomes aligned with Kirkpatrick's 4 levels of evaluation (reaction, learning, behavior, and results) or Miller's Pyramid of clinical competence (knows, knows how, shows how, and does).^14,15 A completed PRISMA-ScR checklist is provided as Supplemental Figure 1.

RESULTS

A total of 3642 records were identified through database searches. After removal of 117 duplicates, 3525 records underwent title and abstract screening. Of these, 115 reports were sought for retrieval, and 106 full-text reports were assessed for eligibility (Figure 1). Six studies from database searching and 10 additional studies from citation searching were included, yielding 16 studies in the final review.

Figure 1.

PRISMA flow diagram.

Using the predefined study classification, 5 studies directly evaluated educational interventions or training outcomes.^16-20 Six studies were training platform studies, and 5 were educationally relevant technology studies.^21-31 The latter group was included to map potentially transferable technology rather than to evaluate those studies as educational trials. A concise synthesis is provided in Table 1, and the full study-level details are provided in Supplemental Table 2.

Table 1.

Study Classification of Included Studies by Educational Relevance

Study classification	Relevant studies	Interpretation
Educational intervention studies (n = 5)	Cinotti et al,16 Su et al,17 Kantar et al,18 Zhang et al,19 Nicot et al20	Direct learner assessment; strongest educational evidence, but outcomes are mainly satisfaction, confidence, or short-term learning
Training platform studies (n = 6)	Kumar et al,21 Hussein et al,22 Zhou et al,23 Ueda et al,24 Powell et al,25 Diaz-Siso et al26	VR/MR, haptic, robotic, cadaveric, or physical platforms for anatomy visualization and rehearsal; technical or feasibility performance is promising, but objective educational validation is limited
Educationally relevant technology studies (n = 5)	Ozmen et al,27 Zhang et al,28,29 Cho et al,30 Suzuki et al31	AI decision support, ultrasound anatomy, and vascular or lymphatic visualization; educational relevance is inferred from transferable training functions and requires prospective learner-centered validation

AI, artificial intelligence; MR, mixed reality; VR, virtual reality.

Across included studies, 15 were in plastic surgery or facial anatomy contexts, whereas 1 was in dermatology. Macro-level imaging was most common (n = 10), followed by meso-level imaging (n = 5) and micro-level imaging (n = 1). Educational outcomes were concentrated around satisfaction, self-reported confidence, perceived anatomical understanding, feasibility, and short-term knowledge. No included study used the Kirkpatrick Model, Miller's Pyramid, or a comparable structured educational framework, and none evaluated long-term retention, supervised transfer into operative practice, or patient-level outcomes.

DISCUSSION

Surgical education has gradually moved from static anatomical resources toward interactive and image-based learning environments. In facial aesthetic surgery training, this transition is particularly important because procedural safety depends on understanding not only surface landmarks but also the depth, variability, and spatial relationships of muscles, vessels, fat compartments, and retaining ligaments. The present scoping review found that available tools are concentrated around visualization and simulation, whereas learner-centered validation remains limited and heterogeneous.

Imaging-Based Visual Learning and Dynamic Anatomical Visualization

Integration of multimodal imaging offers a potentially useful approach to anatomical education. Across point-of-care ultrasound, mixed-reality anatomy, and immersive facial anatomy studies, these tools were used to support spatial awareness and layered anatomical understanding.^21,22,32 We found that visually oriented educational tools more commonly relied on macro- or meso-level multimodal imaging for anatomy training, with a primary investigation emphasis on understanding volume, spatial position, and layered structure. For example, the incorporation of point-of-care ultrasound into anatomy teaching has been reported to increase trainees’ confidence in identifying head and neck anatomical structures and interpreting imaging findings.³² In facial muscle education, Zhang et al proposed a structured “Textbook–Ultrasound–Cadaver” learning model that integrates theoretical anatomy, real-time ultrasound imaging, and cadaveric dissection. This multimodal approach significantly improved learners’ self-reported understanding of facial anatomy and achieved high satisfaction rates.¹⁹ In a related study, ultrasound was shown to enable high-resolution visualization of soft-tissue structures in the face and neck, facilitating precise mapping of age-related anatomical changes.²⁹

Several studies have explored the use of multimodal imaging to reconstruct anatomy-based learning systems. Using high-resolution close-range photogrammetry of cadaveric dissections, Kumar et al constructed a 3D virtual facial model integrated within a mixed-reality platform, enabling detailed visualization of layered facial anatomy, including surface landmarks, vascular structures, nerves, and ligamentous components, while preserving their spatial relationships.²¹ The concept was further expanded by Hussein et al, who integrated high-resolution photogrammetry with virtual reality platforms to construct a reproducible surgical anatomy training environment derived from layered cadaveric dissections.²²

In addition to cadaver-based educational platforms, real-time visual guidance systems may help junior plastic surgeons translate anatomical knowledge into accurate procedural localization. Ultrasound-guided facial botulinum toxin injections have been shown to improve anatomical precision and procedural safety.³³ For example, ultrasound-guided injection of the depressor anguli oris, glabellar muscles, and nasal radix region enables more accurate depth control and may reduce the risk of complications.^28,30 Multimodal visualization techniques have also been combined with emerging imaging modalities. Suzuki et al integrated photoacoustic imaging with AR to generate high-resolution 3D visualization of lymphatic vessels and veins.³¹ This study highlighted how patient-specific vascular or lymphatic mapping could be adapted for preprocedural planning and intraoperative navigation in anatomically complex facial regions. Recent advances in AI-based anatomical recognition further extend the role of multimodal imaging from passive visualization to interactive interpretation.^34-36

More advanced real-time assistance systems further incorporate physical support on the basis of multimodal imaging. Zhou et al developed an AR-assisted surgical system for mandibular angle split osteotomy that integrates AR navigation, robotic manipulation, and haptic feedback.²³ In this system, 3D mandibular models reconstructed from preoperative CT imaging, together with the planned osteotomy pathway, are superimposed onto the surgical field. Drilling and osteotomy can then be performed using a robotic arm equipped with force-feedback sensing, allowing changes in bone density to be monitored and reducing the risk of injury to adjacent neural structures. The technical advancement of surgical systems supports surgeon training with quantitative feedback and outcome measures such as algorithmic accuracy, segmentation consistency, or workflow performance.

Simulation-Based and Haptic Training Platforms

Traditional training models remain valuable but may be limited by restricted availability and reproducibility in some settings. In contrast, virtual simulators can recreate anatomical structures and operative environments in standardized, repeatable formats.^17,21,22,26 Latest surgical systems also incorporate performance feedback and haptic interaction, allowing trainees to practice procedural steps in a controlled setting.¹⁷ In a prospective randomized blinded trial involving 13 plastic surgery residents, Kantar et al demonstrated that a digital cognitive surgical simulator significantly improved learning outcomes for the extended Mohler unilateral cleft lip repair compared with textbook-based instruction.¹⁸ Although cleft lip repair is not a facial aesthetic procedure, this study demonstrated direct evidence that simulation can improve plastic surgery resident learning in a complex facial operation. Similarly, virtual surgical simulators described by Diaz-Siso et al were designed to model craniofacial anatomy and operative workflows, helping trainees integrate anatomical knowledge with procedural decision making.²⁶ These adjacent craniofacial examples inform facial aesthetic education by demonstrating how simulation may support spatial reasoning and procedural rehearsal, but they should not be overinterpreted as proof of benefit in aesthetic procedures without dedicated validation.

Meanwhile, imaging-derived 3D models can be translated into physical simulators for surgical training.²⁵ Using CT- and MRI-based stereolithographic data, Ueda et al developed layered elastic facial models that replicate skin and subcutaneous tissue, enabling realistic simulation of flap design, tissue undermining, and suturing.²⁴ This approach exemplifies the translation of imaging-derived anatomical information into tangible training platforms, bridging the gap between virtual visualization and hands-on surgical simulation.

Language-Model-Assisted Learning: Current Evidence and Limits

Large language models (LLMs) may support surgical education primarily by organizing, explaining, and updating knowledge. For trainees, LLM-based tools can summarize anatomical concepts, translate imaging findings into clinically meaningful explanations, generate self-assessment questions, and provide rapid access to recent literature.^27,37-39 In addition to knowledge acquisition, LLMs may also support text-centered interaction and feedback. Artificial intelligence systems can analyze CT and MRI data alongside trainee-generated reports, providing structured feedback on missed or incorrect findings and tracking longitudinal improvement in image interpretation performance, suggesting a pathway for interns to receive rapid, structured critique of key clinical questions and documentation quality.⁸ In plastic surgery, emerging studies have further explored LLM-based support for intraoperative decision making and postoperative care recommendations, in which natural-language queries can be used to generate structured suggestions regarding procedural choices, risk avoidance, follow-up instructions, and patient counseling.^40,41 However, when used alone, text-based LLMs remain limited by hallucination, incomplete source transparency, and variable domain specificity.⁷ No included study evaluated AI-generated synthetic images as an educational modality; accordingly, image-generation artifacts were not an outcome assessed in this review. In addition, LLM and AI systems may expose or improperly process identifiable patient data. For example, in the US Health Insurance Portability and Accountability Act (HIPAA)-covered settings, when a third-party AI or cloud service creates, receives, maintains, or transmits protected health information on behalf of a covered entity or business associate, an appropriate Business Associate Agreement, where required, is needed in addition to technical cybersecurity safeguards, applicable privacy and data-security requirements, and institutional governance. Overall, their educational value depends on careful verification, linkage to reliable evidence, and integration with expert supervision.

Multimodal AI Agents as Future Learning Systems

Although LLMs primarily support linguistic explanation, knowledge synthesis, and report-based feedback, multimodal AI agents may eventually integrate imaging data, procedural context, and learner-specific performance into adaptive educational workflows. At present, this remains largely a future-facing hypothesis for facial aesthetic surgery education. Studies in medical imaging and adjacent dermatologic applications suggest that AI can guide attention to relevant structures, identify image patterns, and provide structured or explainable outputs.^7,42-45 These examples are relevant only as transferable demonstrations of multimodal image interpretation and feedback, not as direct evidence that AI improves facial aesthetic surgery training. For facial aesthetic education, the plausible role is narrower: comparing trainee-labeled anatomy with expert annotations, flagging uncertainty or high-risk zones, prompting learners to justify procedural decisions, and providing feedback after an independent trainee interpretation. This sequencing is important because AI assistance could also create automation bias or over-reliance if trainees defer to model outputs before developing independent image interpretation and clinical reasoning. Future evaluations should therefore assess not only performance with AI assistance but also independent reasoning and performance after assistance is withdrawn.

Looking ahead, agentic AI systems may extend this paradigm by integrating heterogeneous imaging data into unified, actionable educational platforms.⁴⁶ Future efforts should focus on developing tools that can render multimodal visual and volumetric data in clinically interpretable formats, align multimodal reasoning with current evidence and guidelines, and incorporate verification and self-updating mechanisms for cross-modality anatomical recognition.^47,48 Such systems could enable more personalized and proactive learning, support interactive and adaptive visualization, and strengthen feedback-driven training through analysis of error patterns and high-risk anatomical zones.^38,49 These advances may ultimately improve the efficiency, accessibility, and educational value of surgical training. The proposed integration pathway is summarized conceptually in Figure 2.

Figure 2.

Conceptual framework for integrating multimodal imaging inputs, AI-assisted interpretation, simulation and rehearsal, feedback and learner assessment, and structured educational validation in facial aesthetic surgery training. During the preparation of this work, the authors used ChatGPT (OpenAI, San Francisco, CA) to assist with the generation of this figure. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication. 3D, 3-dimensional; AI, artificial intelligence; AR, augmented reality; CT, computed tomography; VR, virtual reality.

Evidence Gaps and Future Directions for Educational Evaluation

From the perspective of this educational review, many imaging-based and AI-enabled tools were designed and validated as technical or clinical innovations rather than educational interventions. Hence, the educational benefits cannot be directly inferred from current diagnostic accuracy, image segmentation, automated recognition, surgical navigation, or workflow performance data alone. Comparatively fewer studies directly examine how these technologies help early-career aesthetic practitioners acquire anatomical knowledge, develop procedural reasoning, reduce errors, retain skills, or improve supervised clinical performance.

Even among studies with educational outcomes, most report feasibility, anatomical realism, learner satisfaction, confidence, or short-term knowledge acquisition. These outcomes correspond mainly to early educational levels and do not establish durable changes in behavior, operative performance, or patient outcomes. Future studies should therefore incorporate standardized endpoints aligned with the Kirkpatrick Model and Miller's Pyramid, including anatomy recognition accuracy, procedural precision, time to task completion, error rates, confidence calibration, long-term retention, supervised clinical transfer, and downstream patient-safety measures.^14,15 For AI-enabled systems, validation should further include internal and external testing, prospective deployment, bias assessment, explainability, safeguards for expert oversight, and measures of automation bias, over-reliance, and independent performance after AI assistance is withdrawn.

This review is limited by the heterogeneity, fragmentation, and early-stage nature of the available literature, which prevents quantitative pooling or identification of a comprehensive workflow for integrating multimodal imaging into facial aesthetic surgery education. Existing studies largely emphasize technical development or clinical application of modalities such as CT, MRI, ultrasound, 3D photography, and 3D reconstruction, with comparatively limited attention to structured educational implementation, curricular integration, and learner outcomes. Several included studies were from craniofacial and dermatological domains for the shared anatomical focus and learner experience; hence, the direct applicability to facial aesthetic surgery learners and outcomes cannot be assumed. The relatively high proportion of citation-identified studies likely reflects inconsistent indexing of emerging simulation, facial anatomy education, and AI-enabled tools under conventional database education terms. It also raises the possibility that other eligible studies were not captured. Because the database strategy was imaging-centered and did not include a dedicated AI-only term block, AI-focused studies that were not indexed or described in relation to imaging, simulation, or education may have been missed. Citation searching reduced but did not eliminate this risk.

In addition, many emerging AI-related tools remain constrained by limited explainability, incomplete domain-specific validation, variable transparency, and uncertain real-world educational applicability. Therefore, this review was designed as a scoping review; we used descriptive methodological appraisal to identify whether studies evaluated learners, reported objective outcomes, validated their tools, and addressed educational transfer. Consequently, the current evidence base remains insufficient to define best practices for the use of these technologies in training. Nevertheless, the continued expansion of higher-quality imaging datasets, improved validation, and education-focused research suggest substantial potential for multimodal imaging and related digital technologies to strengthen current training approaches through more precise, adaptive, and clinically relevant instruction.

CONCLUSIONS

In summary, multimodal imaging, simulation, and AI-assisted tools may support the evolving educational landscape of facial aesthetic surgery, but the current evidence remains early and heterogeneous. These technologies may support spatial understanding, procedural rehearsal, and individualized feedback, yet few studies have demonstrated objective improvements in operative performance, long-term retention, supervised clinical transfer, or patient outcomes. Future integration should therefore be guided by prospective validation, standardized educational frameworks, compliance with applicable privacy and data-security requirements, safeguards against AI over-reliance, transparent governance, and expert oversight. With these safeguards, imaging-based and AI-assisted systems may become useful components of a modern, evidence-based facial aesthetic surgery curriculum.

Supplemental Material

This article contains supplemental material located online at https://doi.org/10.1093/asjof/ojag189.

Disclosures

The authors declared no potential conflicts of interest with respect to the research, authorship, and publication of this article.

Funding

The work was supported by National Key Research & Development Program of China (2024YFC3405800, 2024YFC3405801), Education and Teaching Reform Project of Graduate in Peking Union Medical College (2024yjsjg001), Peking Union Medical College Hospital “Xingyuan” Program, Peking Union Medical College Hospital Talent Cultivation Program (Category B) No.UGG10110 and Peking Union Medical College Hospital Talent Cultivation Program (Category C) No.UBJ11557. The views expressed in this manuscript are solely those of the authors and do not necessarily represent the official positions of their affiliated institutions or funding organizations.