Integrated AI-Based Facial Image Analysis for Activity and Severity Assessment in Thyroid Eye Disease: Validation of Glandy CAS, EXO, and LID
Author: Antonio Manuel Garrido Hermosilla
Base Hospital / Institution: Virgen Macarena University Hospital (Seville, Spain)
ePoster presentation
Abstract ID: 25-259
Purpose
Glandy CAS, EXO, and LID are artificial intelligence (AI)-based software devices developed to assess thyroid eye disease (TED) using standard facial photographs. Glandy CAS estimates the Clinical Activity Score (CAS) to identify active TED, while Glandy EXO measures proptosis and Glandy LID evaluates eyelid retraction and ocular surface exposure. This study aimed to validate the clinical performance of these systems.
Methods
Glandy CAS was validated through a confirmatory clinical trial involving 756 facial photographs. The system’s ability to detect active TED (CAS ≥ 3) was compared to general ophthalmologists’ CAS scoring, using in-person assessments by an oculoplastic specialist as the reference standard. Glandy EXO was trained and internally validated on 1,610 DSLR images from 1,108 TED patients and externally validated using 678 images from 171 patients. Glandy LID assessed midpupil lid distance (MPLD) and ocular surface area, and its outputs were compared with expert manual measurements.
Results
Glandy CAS achieved a sensitivity of 87.9%, specificity of 95.8%, and an F1 score of 0.88, outperforming general ophthalmologists (F1: 0.57). It predicted CAS within one point of the reference in 82.3% of cases. Glandy EXO demonstrated high agreement with exophthalmometry, with Pearson correlations of 0.82 (internal) and 0.77–0.79 (external), and mean absolute errors (MAE) ranging from 1.23 to 1.27 mm. Glandy LID achieved excellent correlation with manual measurements: for MRD1, Pearson r = 0.94 (MAPE 9.53%); for MRD2, r = 0.91 (MAPE 14.63%). Ocular surface area estimates also closely aligned with reference annotations.
Conclusion
The Glandy AI suite enables accurate, noninvasive assessment of TED activity and severity through facial image analysis. The three systems demonstrated strong agreement with clinical reference standards across diverse datasets, supporting their integration into routine care for early detection, longitudinal monitoring, and accessible evaluation of TED.
Additional Authors
| First name | Last name | Base Hospital / Institution |
|---|---|---|
| Kyubo | Shin | Thyroscope Inc. (Seoul, Republic of Korea) |
| Marina | Soto Sierra | Virgen Macarena University Hospital (Seville, Spain) |
| Raquel | Monge Carmona | Virgen Macarena University Hospital (Seville, Spain) |
| Mariola | Méndez Muros | Virgen Macarena University Hospital (Seville, Spain) |
| Jae Hoon | Moon | Thyroscope Inc. (Seoul, Republic of Korea) |
| Jongchan | Kim | Thyroscope Inc. (Seoul, Republic of Korea) |
| Joonhyeon | Park | Thyroscope Inc. (Seoul, Republic of Korea) |
