Artificial intelligence in oculoplasty: Current applications and future perspectives

Abstract

Artificial intelligence (AI) is rapidly transforming clinical ophthalmology, with most validated applications in retinal and corneal imaging. In contrast, oculoplastic surgery – which encompasses eyelid, orbital, lacrimal, and periocular disorders – has only recently begun to incorporate AI into clinical workflows. Emerging applications include automated image analysis for eyelid malpositions, deep learning-based interpretation of orbital imaging, AI-assisted planning for dacryocystorhinostomy, and predictive modeling in reconstructive and aesthetic procedures. Despite encouraging results, challenges such as limited datasets, imaging heterogeneity, lack of multimodal integration, and regulatory readiness continue to hinder widespread clinical adoption. This mini-review summarizes current AI applications in oculoplastic surgery, identifies key knowledge gaps, and discusses future directions toward personalized, multimodal, AI-driven surgical care.

Key Words: Artificial intelligence; Machine learning; Deep learning; Oculoplastic surgery; Periocular imaging; Thyroid eye disease; Lacrimal surgery

Core Tip: Artificial intelligence (AI) is rapidly transforming the landscape of oculoplasty, extending its impact beyond retinal and corneal imaging into eyelid, orbital, lacrimal, and reconstructive surgery. AI-driven systems now automate periocular measurements, detect and classify orbital tumors, assist in dacryocystorhinostomy planning, and predict postoperative aesthetic outcomes, enabling greater precision, consistency, and efficiency. However, clinical adoption remains limited by small, heterogeneous datasets, imaging variability, and ethical or regulatory challenges. Continued collaboration between clinicians, data scientists, and engineers is essential to develop validated, transparent, and user-friendly tools. As multimodal AI platforms, robotic-assisted surgery, and tele-oculoplasty advance, they promise to usher in a new era of personalized, data-driven oculoplastic care – seamlessly combining surgical expertise with intelligent technology to enhance both functional and aesthetic outcomes.

INTRODUCTION

Artificial intelligence (AI) has made substantial progress in medical imaging, particularly in radiology and ophthalmology. While retinal image analysis dominates AI-based ophthalmic research, oculoplasty – comprising functional and aesthetic surgery of the eyelids, orbit, lacrimal system, and periocular region – remains a relatively underexplored domain. The complexity of periocular anatomy, the heterogeneity of disease presentations, and the reliance on imaging make oculoplasty uniquely suited to AI applications.

AI has emerged as a transformative frontier in ophthalmology, capable of simulating human cognitive processes such as visual recognition, pattern interpretation, and decision-making through complex computational models[1-3]. Its clinical impact has already been established in retinal imaging, diabetic retinopathy screening, glaucoma detection, and age-related macular degeneration grading[2,3]. These advances have inspired oculoplasty surgeons to explore the application of AI in functional precision and aesthetic judgement in eyelid and orbital conditions.

Recent reviews and cross-sectional studies have emphasized the emerging applications and the growing awareness of AI among oculoplasty surgeons. A survey-based study by Fazekas et al[4] explored provider perspectives on AI integration in oculoplastics. It highlighted both enthusiasm for its potential and concern over data standardization, validation, and ethical governance[4]. The study underscored that while most practitioners recognize the promise of AI for diagnostic imaging and periocular morphometry, significant gaps remain in clinical translation and regulatory guidance.

Oculoplasty presents a unique intersection of imaging, anatomical variability, and individualized planning, making it ideally suited for AI-driven solutions[5]. In this review, the term AI is used as an umbrella concept encompassing traditional machine learning, deep learning, and computer vision techniques. Deep learning refers to multilayer neural networks capable of automated feature extraction, while explainable AI aims to enhance the interpretability of model outputs. Multimodal AI integrates heterogeneous data sources, including imaging and clinical variables, to provide more comprehensive decision support.

Recent developments have expanded the use of AI in this domain. Automated periocular morphometry has enabled objective grading of eyelid malpositions, such as ptosis, dermatochalasis, and lid retraction[6]. Deep learning applied to orbital imaging can detect, segment, and classify orbital pathologies, including tumours and thyroid eye disease (TED)[7,8]. Deep learning-accelerated image reconstruction in orbital magnetic resonance imaging (MRI) significantly shortens acquisition time while enhancing image quality, thereby improving patient comfort and diagnostic reliability[9]. In the lacrimal system, AI has been used for endoscopic image interpretation and procedural guidance in dacryocystorhinostomy (DCR)[10]. Moreover, in plastic and reconstructive surgery, AI-driven facial analysis enables preoperative simulation and postoperative outcome prediction[11].

Despite these promising innovations, the field faces challenges. Compared to retinal imaging, oculoplastic AI research is limited by smaller datasets, variable imaging modalities [clinical photographs, computed tomography (CT), MRI, endoscopy], and ethical considerations such as data privacy and algorithmic transparency. A recent comprehensive review by Ing and Bondok[12], summarized the evolving role of AI in oculoplastic practice, emphasizing its applications in diagnostic imaging, periocular morphometry, and surgical planning, while identifying data scarcity and validation gaps as key barriers.

Future AI systems in oculoplasty are expected to evolve beyond single-task algorithms toward multimodal frameworks that integrate clinical photographs, radiologic imaging, demographic variables, and electronic health records. Such models enable personalized risk stratification, outcome prediction, and treatment selection by combining structural, functional, and clinical data. However, real-world implementation faces practical challenges related to image quality variability. Differences in camera systems, lighting conditions, imaging protocols, and operator expertise can degrade AI performance. Strategies to mitigate these issues include image-quality-aware training, pre-processing pipelines, domain adaptation, and federated learning approaches that allow model training across institutions without centralized data sharing. Addressing these challenges is essential for equitable and scalable AI deployment in oculoplastic practice.

In this context, the present mini-review aims to synthesize recent advances in AI applications across key oculoplastic domains, critically evaluate their methodological robustness, and identify barriers to real-world clinical translation. Particular emphasis is placed on multimodal AI integration, generalizability across imaging environments, and regulatory and ethical considerations that will shape the future adoption of AI in oculoplastic practice.

APPLICATIONS OF AI IN OCULOPLASTY

AI in eyelid disorders

AI is rapidly improving how clinicians quantify and manage eyelid disorders by automating periocular biometrics, aiding diagnosis, and enhancing counselling. Independent groups have demonstrated automated pre-operative/post-operative eyelid morphometry around ptosis surgery using deep learning, reducing observer bias in outcome reporting[13]. AI has also been used to predict postoperative appearance after blepharoptosis repair, improving expectation-setting and shared decision-making[14]. For case finding and triage, convolutional neural networks (CNNs) can detect blepharoptosis from facial images under realistic conditions, offering scalable screening adjuncts[15]. Deep-learning landmark detectors now extract standardized measurements, such as margin reflex distance (MRD) 1/MRD2, palpebral aperture, and brow position, directly from photographs, yielding strong agreement with clinician measurements and enabling reproducible longitudinal audits[16]. Beyond clinic cameras, smartphone-based pipelines can estimate MRD1/MRD2 and levator function with excellent correlation with gold standards, supporting point-of-care and telemedicine workflows[17]. Expanding beyond static photos, video-based analysis can quantify blink frequency and completeness in facial palsy/synkinesis, offering objective functional endpoints relevant to exposure risk and rehabilitation[18]. Expanding on future directions, infrared imaging combined with deep learning is emerging as a promising approach for measuring MRD1 under standardized lighting conditions. This approach has shown excellent agreement with traditional manual measurements while offering consistent image quality across varied clinical environments. It broadens the possibilities for accurate, contact-free eyelid assessment and could make periocular evaluation more accessible and reliable in both clinical and remote settings[19]. While AI applications in eyelid disorders primarily focus on surface-level morphometry, deeper anatomical and volumetric challenges are addressed in orbital pathologies.

AI in orbital disorders

AI is steadily advancing the understanding and management of orbital disorders, from tumour characterization to TED and surgical planning. Deep learning applied to orbital CT and MRI can now distinguish benign from malignant orbital tumours, predict growth patterns, and delineate critical anatomical boundaries essential for surgery. Shao et al[19] developed a fully automated deep learning pipeline that segments orbital masses and classifies lesion type with diagnostic accuracy comparable to that of experienced radiologists. Similarly, Nakagawa et al[20] showed that CNNs can reliably detect orbital invasion by sinonasal tumors, thereby improving sensitivity and assisting in surgical planning. In parallel, AI-based segmentation of orbital bones and soft tissues enables rapid three-dimensional reconstruction for volumetric assessment and personalized preoperative planning; Hamwood et al[21] demonstrated a model capable of precise orbital boundary mapping across CT and MRI modalities. In TED, AI tools now quantify extraocular muscle and orbital fat volumes, helping to grade disease activity and predict treatment outcomes. A recent study by Alkhadrawi et al[22] used deep learning-based segmentation to assess muscle and fat compartments, achieving high accuracy in identifying severe disease and optic neuropathy. AI is also transforming how clinicians monitor exophthalmos – Park et al[23] introduced a deep learning system that estimates proptosis directly from facial photographs, showing a strong correlation with conventional Hertel exophthalmometry.

Meanwhile, Han et al[24] demonstrated a segmentation-based model that detects eyelid retraction, ocular surface inflammation, and movement abnormalities in TED with over 94% accuracy. Together, these developments illustrate how AI is enhancing diagnostic precision, streamlining surgical workflows, and paving the way toward personalized, image-guided management of orbital diseases. Beyond orbital imaging and disease classification, AI is increasingly being explored for the functional assessment of complex lacrimal disorders.

AI in lacrimal surgery

AI is increasingly influencing the diagnosis and surgical management of lacrimal drainage disorders, offering tools for greater accuracy and efficiency. Deep learning applied to dacryocystography has demonstrated expert-level detection and localization of nasolacrimal duct obstruction, significantly reducing interpretation time and operator dependence[10]. AI models trained on anterior-segment OCT can differentiate normal tear meniscus patterns from those seen in obstruction, providing a potential non-invasive triage tool before invasive testing[25]. In the endoscopic domain, CNNs applied to nasal and dacryoendoscopic images have enabled automated identification of mucosal inflammation, debris, and obstructive lesions, assisting real-time intraoperative decision-making[26].

AI has also improved preoperative planning. CT-based segmentation models, such as nnU-Net v2, can accurately delineate the nasolacrimal canal, aiding visualization and pathway mapping for endonasal DCR[27]. Predictive algorithms have been used to anticipate the need for concomitant septoplasty during endoscopic DCR, enabling better operative preparation[28]. Furthermore, mixed-reality navigation combining CT data with AI-guided endoscopic mapping has improved surgical localization of the lacrimal sac and selection of the osteotomy site in complex cases[29]. Collectively, these advances demonstrate how AI – spanning diagnostic imaging, intraoperative guidance, and surgical planning – enhances precision, reproducibility, and safety in lacrimal surgery, marking an important step toward data-driven oculoplastic care.

Despite promising results, most AI studies in lacrimal surgery are limited by small, single-centre datasets and lack prospective or external validation. Additionally, variability in endoscopic image quality and surgical technique poses challenges for algorithm generalizability.

These developments naturally extend into aesthetic and reconstructive oculoplasty, where outcome prediction and personalization are central.

AI in aesthetic and reconstructive surgery

AI is steadily becoming a practical ally in aesthetic and reconstructive oculoplasty – making measurements objective, predictions personalized, and reconstructions more precise. On the assessment side, deep-learning image analysis can now quantify eyelid metrics (e.g., MRD1/MRD2, contour symmetry) before and after ptosis surgery, with good agreement with clinicians, reducing observer bias and enabling consistent longitudinal audits[30]. For counselling and expectation-setting, a fully automatic postoperative appearance prediction system uses pre-op photos to generate realistic post-operative simulations for blepharoptosis, showing sub-millimetre MRD1 errors in most cases and high patient/expert satisfaction[31]. In reconstruction, CT-based deep learning can segment the orbit with clinically useful accuracy to speed planning and guide patient-specific implant design, and thin-wall-aware networks further improve orbital wall segmentation for pre-op modelling and implant fabrication[32,33]. Complementing these pipelines, classification models trained on periocular images can reliably flag ptosis – useful for triage and teleophthalmology, and as a front-end quality gate for morphometric tools[34]. Many aesthetic AI models rely on retrospective photographic datasets and subjective outcome metrics, raising concerns about bias, overfitting, and real-world applicability across diverse populations. A schematic overview of current AI applications across eyelid, orbital, lacrimal, and reconstructive oculoplasty domains is presented in Figure 1, summarizing the workflow from data input to AI-assisted analysis, surgical planning, and outcome prediction. The clinical relevance and limitations have been summarized in Table 1, which highlights the comparative progress across domains.

Figure 1 Conceptual summary of artificial intelligence applications in oculoplasty. The diagram illustrates how artificial intelligence (AI) integrates across four key domains – eyelid, orbital, lacrimal, and reconstructive surgery – each contributing to diagnostic precision, surgical planning, and outcome prediction. A central AI core connects these domains through a workflow of data input: (1) AI analysis; (2) Surgical planning; and (3) Outcome prediction. Surrounding elements highlight current challenges (data scarcity, ethical and medico-legal concerns, and clinical integration) and emerging trends shaping the field’s future (multimodal AI, robotic navigation, and tele-oculoplasty). AI: Artificial intelligence; DCR: Dacryocystorhinostomy.

Table 1 Summary of recent artificial intelligence applications in oculoplastic surgery.

Domain	Clinical task	Data modality	AI technique	Clinical relevance/limitations
Eyelid disorders	Ptosis grading, eyelid morphometry	Clinical photographs	CNN, U-Net	Objective measurement; limited by retrospective, single-centre datasets
Orbital pathologies	Tumor classification, TED assessment	CT, MRI	Deep learning segmentation, radiomics	Improved surgical planning; limited external validation
TED	Muscle/fat volume quantification	CT, MRI, facial images	U-Net, CNN	Severity grading; heterogeneous imaging protocols
Lacrimal disorders	Nasolacrimal duct obstruction detection, dacryocystorhinostomy planning	Dacryocystography, endoscopy	CNN, image classification	Intraoperative guidance; small sample sizes
Aesthetic and reconstructive	Outcome prediction, facial symmetry	Facial photographs, three-dimensional imaging	Generative AI, morphometric models	Patient counselling; risk of bias and overfitting

AI: Artificial intelligence; CNN: Convolutional neural network; CT: Computed tomography; MRI: Magnetic resonance imaging; TED: Thyroid eye disease.

CHALLENGES AND LIMITATIONS

Although reported accuracies are encouraging, most AI studies in oculoplasty remain retrospective, single-centre investigations with limited sample sizes. External validation, prospective testing, and comparison against expert consensus are frequently lacking. These limitations restrict generalizability and underscore the need for standardized datasets and multicentre collaboration. Even though AI has made impressive strides in periocular and lacrimal surgery, its journey from research to routine clinical use is still fraught with challenges. One of the biggest hurdles is data scarcity – unlike retinal imaging, which benefits from large, well-annotated databases such as EyePACS and Messidor, periocular and lacrimal imaging data remain limited, inconsistent, and often poorly standardized. This lack of high-quality data makes it difficult for AI models to learn robust patterns and perform reliably across different patient populations.

Additionally, the wide variety of input sources – ranging from clinical photographs to CT, MRI, and endoscopic images – creates inconsistencies that complicate model training and validation across institutions. Ethical and legal concerns also loom large: (1) Questions about patient privacy; (2) The transparency of AI decision-making; and (3) Medico-legal responsibility in the event of diagnostic or surgical errors remain unresolved. Finally, even when effective algorithms exist, clinical integration poses practical barriers. The absence of intuitive, user-friendly platforms, limited compatibility with hospital information systems, and differing levels of trust or acceptance among clinicians all slow down the adoption of AI tools in everyday periocular practice. Ethical and regulatory considerations are central to the deployment of AI in oculoplasty. Most AI tools remain investigational and are not yet approved under regulatory frameworks such as Food and Drug Administration clearance or Conformite Europeenne marking. Dataset governance, patient consent, and algorithmic bias – particularly in periocular imaging across ethnicities – remain unresolved concerns. Transparent reporting, bias auditing, and regulatory alignment are essential before clinical integration.

FUTURE DIRECTIONS

Looking ahead, AI is poised to reshape periocular and lacrimal surgery in exciting ways. The next generation of AI tools will move beyond single-task algorithms toward multimodal systems that combine imaging, clinical details, and electronic health records to offer richer, more personalized decision support. Advances in AI-guided surgical navigation and robotics could bring a new level of precision to delicate procedures such as endonasal DCR and complex orbital reconstructions, helping surgeons plan and execute interventions with greater confidence. The concept of tele-oculoplasty – where AI assists in triaging images and guiding remote consultations – holds promise for expanding access to specialist care, particularly for patients in underserved or rural areas, while also improving postoperative monitoring. Finally, integrating personalized predictive analytics could enable surgeons to forecast surgical outcomes and tailor interventions to each patient’s anatomy and risk profile, marking a shift toward truly data-driven, precision-based oculoplasty.

CONCLUSION

Although still in its early stages, AI shows remarkable potential to transform the field of oculoplasty – from improving diagnostic accuracy to guiding surgical planning and predicting outcomes with greater precision. Realizing this potential will require close collaboration between ophthalmologists, radiologists, and computer scientists to design AI systems that are not only technically robust but also clinically meaningful and ethically responsible. Moving forward, research should prioritize the creation of large, multicentre datasets and the standardization of imaging protocols, ensuring that AI models learn from diverse, high-quality data. Equally important is the establishment of clear regulatory and validation frameworks to guarantee that these technologies can be safely and effectively integrated into everyday oculoplastic practice, ultimately enhancing patient care and surgical outcomes.