Periorbital Measurements – Automated versus Semi-Automated and Manual Approaches

Abstract

Purpose : An artificial intelligence algorithm, OrbitMap, was created for automating periorbital measurement using pre-developed Google MediaPipe Facemesh, along with a semi-automated ImageJ tool.

Methods : Craniofacial front-facing cleft palate images were included. Five human graders conducted measurements manually and with the semi-automated approach. Measurements included margin reflex distances (MRD1, MRD2), medial intercanthal distance (ICD), inner pupillary distance (IPD), outer canthal distance (OCD), vertical dystopia (VD), canthal tilt (CT), inferior scleral show (ISS), and medial, central, and lateral brow height (MBH, CBH, LBH). Intraclass correlation coefficients and mean difference between methods were calculated.

Results : 45 photos were analyzed. Mean difference between manual and semi-automated was <0.9 mm except for MBH, LBH, and IPD, with statistical reliability between the methods for all measures. Between semi-automated versus automated, and manual versus automated, all measures except VD and ISS had mean difference >0.9mm, with statistical reliability between all methods except ISS (p>0.05). Average processing time/image was 38.9±14.8 minutes for manual measurements, 6.8±3.0 min for the semi-automated ImageJ macro and 3.2s for automated OrbitMap (p < 0.001).

Conclusions : Significant reliability exists between the methods for most facial measurements, with semi-automated and manual measurements.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.