Marker-Free, Automated Eyelid Assessment in Thyroid Eye Disease Using Artificial Intelligence: A Multicenter...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Thyroid eye disease (TED) management requires accurate eyelid morphology assessment.
• Current methods for measuring eyelid parameters like margin reflex distance (MRD1, MRD2) can be subjective or require physical markers.
• Automated, objective, and marker-free solutions are needed for consistent TED monitoring.

Purpose of the Study:

• To validate Glandy LID, a novel deep learning-based software for automated eyelid morphology assessment.
• To evaluate the software's accuracy and generalizability using intrinsic corneal diameter for calibration.
• To establish Glandy LID as a reliable, marker-free tool for clinical use.

Main Methods:

• A multicenter retrospective study involving two cohorts (n=119 and n=140) with TED.
• An AI algorithm segmented eyelid and corneal regions from facial photographs (DSLR and smartphone images).
• Eyelid metrics were calculated using corneal diameter for scale; internal validation used physical markers, external validation used clinical records.

Main Results:

• Internal validation showed high geometric precision (IoU 0.94) and excellent agreement for MRD1 (PCC 0.98, MAPE 5.69%) and MRD2 (PCC 0.94, MAPE 4.57%).
• External validation with smartphone images demonstrated strong reliability for MRD1 (PCC 0.94, MAPE 9.06%) and MRD2 (PCC 0.93, MAPE 15.07%).
• The AI system maintained high accuracy despite variations in image source and measurement comparison methods.

Conclusions:

• Glandy LID provides an accurate, robust, and automated solution for eyelid measurement without physical markers.
• The software overcomes limitations of manual and marker-based assessments, offering a practical tool for TED monitoring.
• This AI-driven approach enhances accessibility for objective TED assessment in routine and remote healthcare.