EyeModelHub
Home Subspecialties Oculoplastics Model card
§Model Card

Periocular Measurement AI Platform (HRNetV2)

Automated calculation of periocular measurements (MRD1, MRD2, PFH, IICD, OICD, IPD, HPA) from standardized facial photographs

Rana K. · 2024 · Australia · doi:10.1177/11206721241249773
Open DOI
Measurement/Regression CNN External photo Internally validated
More attributes
Anatomy · Periorbital soft tissueConditions · Periorbital measurementSingle-centern = 479 patients · 958 eyes images
§01Overview

Brief overview

This study developed an HRNet-v2 deep learning facial landmark detection network to automate periocular anthropometric measurements from standardized facial photographs, including MRD1, MRD2, palpebral fissure height, interpupillary distance, and canthal distances. Across 958 eyes from 479 prospectively enrolled patients at a single Australian center, the model achieved mean absolute errors below 1 mm for all measurements and excellent agreement with human graders (ICC >0.90 for most metrics). The system used manually annotated periocular landmarks and heatmap-based landmark localization to derive linear measurements. No external validation, public code, pretrained weights, or clinical deployment were reported. The model was evaluated under standardized imaging conditions and remains internally validated for research use.

§02Performance
Headline performance

MAE (mm): MRD1 0.29, MRD2 0.30, PFH 0.31, HPA 0.74, OICD 0.73, IICD 0.88, IPD 0.22.

Full metrics — as reported
MAE (mm): MRD1 0.29, MRD2 0.30, PFH 0.31, HPA 0.74, OICD 0.73, IICD 0.88, IPD 0.22.
ICC: 0.90–0.99 across metrics; landmark error rate 0.51% with no failures.
Calibration · Uncertainty
Not reported.
Bias · Fairness
Not reported.
§03Data & Validation
Study designDiagnostic accuracy study
Center typeSingle-center
Patients479
Images / eyes958 eyes
Split70/30
Reporting frameworks None stated
Age rangeMean 59
Sex (% female)Not reported
Race / ethnicity reportedN
Subgroup performanceN
§04Deployment
Deployment maturity
  1. Concept
  2. Prototype
  3. Internal val.
  4. External val.
  5. Regulator
  6. Clinical
Regulatory status
None No FDA / Health Canada / CE clearance.
Artifacts & license
Code repository none
Pretrained weights none
Live demo none
License Not specified
§05Quality Review · APPRAISE-AI
Total score
37 /100 Low
ClinicalDataMethodRobust.Report.Repro.
Sub-domain scores
Clinical relevance 4/4 100%
Data quality 8/24 33%
Methodological conduct 4/20 20%
Robustness of results 2/20 10%
Reporting quality 12/12 100%
Reproducibility 7/20 35%
Why this score?

The APPRAISE-AI tool grades AI/ML medical studies across 24 items in 6 subdomains, scored independently by two reviewers. The total reflects the methodological transparency of the publication — not the underlying clinical value of the model.

How to read this: high scores in clinical relevance and reporting with low scores in methodological conduct, robustness, or reproducibility is the dominant pattern in ophthalmic AI today.

Quality bands: Low <40 · Moderate 40–59 · High 60–79 · Very High 80+. Most published ophthalmic AI sits in the Low–Moderate band; studies that break into High typically combine multi-center datasets, prospective evaluation, and open-source release.

§06Full Metadata
Citation & identification
Bibliographic record
Model IDoculoplastics_rana_2024
Full author listBeecher, M., Caltabiano, C., Macri, C., Zhao, Y., Verjans, J., & Selva, D.
JournalEur J Ophthalmol
DOI10.1177/11206721241249773
ConditionsPeriorbital measurement
Architecture
Model internals
Architecture familyCNN
Architecture detailHRNet-v2 landmark detection model.
Funding & conflicts
Disclosures
FundingNot reported
Conflicts of interestNo conflicts
Card lifecycle
Status, version, reviewers
Card statusPublished
Card versionv1.0.0
First added2026-04-28
Last reviewed2026-04-28
Reviewersattribution opt-in pending
Update history
Card changelog
  • 2026-04-28 initial publication (v1.0.0)
← Back to Oculoplastics