Oculoplastics
AI/ML in eyelid, orbit, lacrimal, and periorbital aesthetic care.
24 models appraised.
Sorted by APPRAISE-AI total. Tap any card for the full appraisal, methodology notes, and reviewer comments.
Post-Blepharoplasty Age Prediction AI (FaceAge)
Best individual model: Face++ (ICC 0.78; MAE 4.82 yrs).
If you want AI-based estimation of perceived rejuvenation after blepharoplasty, FaceAge offers a reproducible research framework using commercial age-prediction APIs — but it remains an exploratory cosmetic-outcomes tool, not a validated clinical assessment system.
PeriorbitAI
Dice coefficients 0.82–0.96; mean absolute error ~0.5–2 mm vs human graders; ICC 0.90–0.95.
If you need automated MRD1/MRD2 from a clinic photo, this is the most reproducible openly-available option — but treat it as a research aid, not a measurement device.
Facial Age & Attractiveness AI Platform
Age model categorical accuracy ≈51% on an internal test set.
If you need exploratory AI-based facial age estimation after orthognathic surgery, this model demonstrates feasibility — but limited accuracy, absent external validation, and no clinical deployment make it a research-only tool.
Cornea Exposure Rate AI System (AnigmaView)
AI vs manual ImageJ: ICC 0.98–0.99 (excellent).
Can be considered as an adjunct. AI-based CER measurement shows near-perfect agreement with manual ImageJ while enabling faster, standardized assessment of ptosis surgical outcomes, but should not replace clinical judgment or comprehensive eyelid evaluation.
Orbital Volume Segmentation AI (Human-in-the-Loop U-Net)
Manual-only dataset (178 CTs): Dice ≈ 0.90.
If you need rapid orbital volume segmentation from CT, this AI-assisted U-Net workflow appears faster than manual annotation with similar accuracy — but it remains a single-centre research tool without external validation or clinical deployment.
Strabismus Surgical Planning AI Platform
Screening: AUC 1.00 (retrospective) / 0.98 (prospective); accuracy 99% / 97%.
If you triage suspected strabismus from photos, this is the most prospectively-validated reference model in the literature. Surgical-target-angle prediction is preliminary and not yet a substitute for the bedside exam.
Orbital Segmentation AI Model (2D U-Net)
Overall Dice = 0.86; ASSD = 0.71 mm.
If you need rapid automated orbital CT segmentation for fracture planning and 3D printing, this is a promising openly available research tool — but it remains internally validated and should not replace expert surgical review.
Blepharoptosis AI Platform (DenseNet121)
Best model (DenseNet121, no pretraining): sensitivity 0.90; specificity 0.82; accuracy 0.89; AUROC 0.95.
If you need automated ptosis screening from a standard clinic photograph, this CNN-based approach shows strong internal diagnostic performance — but it remains an early research tool without external validation or real-world clinical deployment.
Referable Blepharoptosis AI (VGG16 CNN)
On a 50-image test set: accuracy 0.90; sensitivity 0.92; specificity 0.88; AUC 0.99. Model outperformed three non-ophthalmic physicians (mea
If you need automated screening for referable ptosis from periocular photographs, this CNN model outperformed non-ophthalmic physicians — but it remains an internally validated referral aid, not a standalone diagnostic tool.
Orbital Fracture AI Platform (InceptionV3-XGBoost)
Slice-level accuracy = 0.92; patient-level accuracy = 0.87; patient-level AUC = 0.96.
If you need automated detection of orbital blowout fractures on CT, this CNN–XGBoost pipeline shows strong internal performance but remains a single-centre research tool. Use it as a proof-of-concept decision support system, not a clinically deployable diagnostic device.
Orbital Fracture AI Platform (DenseNet169-UNet)
AUC 0.99; accuracy 0.97 (classification); Dice 0.88 (segmentation).
If you need automated orbital fracture detection and segmentation on CT, this DenseNet-169 + U-Net model is highly accurate in internal testing, but it remains a single-centre research tool without external validation, code release, or clinical deployment.
Lacrimal Obstruction AI Platform (AS-OCT CNN)
Best single model (DenseNet-169): AUC 0.78; sensitivity 64.6%; specificity 72.1%.
If screening AS-OCT for lacrimal duct obstruction, this ensemble CNN shows moderate diagnostic accuracy, but remains a single-center research model with no external validation or deployment—best used as an exploratory tool rather than a clinical decision system.
Blepharoptosis Morphology Analysis AI (Attention R2U-Net)
Segmentation Dice: eyelid 0.96, cornea 0.96.
If you need automated periocular measurements from standard clinic photographs, this model shows highly reproducible MRD and eyelid contour analysis — but it remains a single-centre research tool without external validation or clinical deployment.
Smartphone Eyelid Measurement AI (MAIA)
MRD1: MAE 0.35 mm; ICC 0.90.
If you need rapid smartphone-based MRD1/MRD2 assessment, this model shows good agreement with clinician measurements — but it remains a single-center research prototype without external validation or clinical deployment.
Blepharoptosis AI Platform (Xception CNN)
Baseline model: AUC 0.87; sensitivity 0.68; specificity 0.89; accuracy ~90%.
If you need AI-based screening for functionally significant ptosis from periocular photos, this model shows promising sensitivity — but it remains a single-center research tool without external validation or clinical deployment.
Orbital Wall Segmentation AI (Thin-WallNet)
WOW: Dice 96.1%, IoU 92.5%, 95HD 0.51 mm.
If you need rapid automated orbital wall segmentation from CT for reconstruction planning, this model shows strong internal accuracy — but it remains a single-center research tool without external validation or clinical deployment.
Periocular Measurement AI Platform (HRNetV2)
MAE (mm): MRD1 0.29, MRD2 0.30, PFH 0.31, HPA 0.74, OICD 0.73, IICD 0.88, IPD 0.22.
If you need automated periocular measurements from standardized clinic photographs, this HRNet-based system shows strong agreement with human graders — but it remains an internally validated research tool without external or real-world telehealth validation.
Abnormal Lid Position Assessment AI (OpenFace)
Automated vs manual VPA/IPD: r = 0.87 (r² = 0.76); 94% of values within Bland–Altman limits of agreement. Robust to wide variation in image
If you need low-cost automated palpebral aperture assessment from routine facial photographs or webcam video, OpenFace is a transparent and reproducible research option — but it remains a feasibility-stage tool, not a validated clinical measurement system.
iOS Blepharoptosis Detection AI (MobileNetV2)
5-fold cross-validation: AUC 0.90; accuracy 0.83; sensitivity 0.83; specificity 0.83.
If you need rapid automated ptosis screening from tablet photographs, this MobileNetV2-based tool shows reasonable diagnostic accuracy — but it remains a single-centre research prototype without external validation or clinical deployment.
Eyelid Morphometry AI Platform (DeepLabV3+)
ICC vs manual measurement: MRD1 = 0.97; MRD2 = 0.94; upper length = 0.80; lower length = 0.74; narrow Bland–Altman limits of agreement. Auto
If you need automated MRD1/MRD2 and eyelid contour analysis from standardized facial photographs, this system shows strong agreement with manual grading — but it remains a single-center research tool without external clinical validation.
KestyAI
Reported cohort averages only (no statistical testing):
If you need AI-based cosmetic grading from periocular photographs, Kesty AI shows potential as a standardized research assessment tool — but the closed model, absent validation metrics, and lack of reproducibility limit clinical interpretability.
Orbital Trauma Planning Platform (Mimics-Romexis)
Segmentation time ~10 min (AI) vs ~63 min (semi-automated); DSC 0.92–0.93; Jaccard 0.88–0.89; Hausdorff Distance ~0.92 mm; expert clinical utility score 4.4/5.
Two commercial, FDA-cleared tools can produce orbital fracture segmentations of clinical utility in roughly 10 minutes per case — a credible alternative to manual semi-automated planning.
Periorbital Rejuvenation Analysis AI Platform
Platforms generally underestimated true age; preoperative percentage errors ranged 4–18%.
If you want an objective estimate of perceived rejuvenation after periorbital surgery, CNN-based facial age analysis can quantify small postoperative age changes — but it remains a retrospective research tool, not a validated clinical outcome measure.
Preoperative Counselling Diffusion Model (DALL·E2)
No quantitative evaluation. Satisfactory images required 1–3 attempts; end-to-end workflow >3 minutes per case.
If you want rapid AI-generated visualizations for oculoplastic preoperative counselling, DALL·E 2 shows early promise — but treat it as a speculative visualization tool, not a reliable predictor of surgical outcomes.
Try widening your selection or .