Clinical Validation of AI-Based Quantitative Eye Movement Analysis From Smartphone 9-Gaze Videos (Glandy EOM)

Ocular motility is commonly evaluated using the 9-gaze examination, which relies on clinician observation and semi-quantitative grading. This approach is inherently subjective and can show inter-observer variability and limited sensitivity to subtle changes over time. Objective, quantitative assessment of eye movement remains an unmet clinical need, particularly for monitoring disease progression and treatment response.

Recent advances in artificial intelligence have enabled automated analysis of facial and ocular features from smartphone-recorded videos, offering the potential for reproducible quantitative measurement of eye movement. Glandy EOM is an investigational software-only medical device (SaMD) developed by THYROSCOPE INC. that processes smartphone-recorded 9-gaze videos to generate quantitative eye movement metrics. Glandy EOM has not yet received regulatory approval and is used for research purposes only in this study.

The study is a prospective, multicenter, multinational observational investigation designed to evaluate the clinical validity of Glandy EOM. The study population consists of adults with suspected or confirmed ocular motility abnormalities; healthy volunteers without any clinical indication for 9-gaze examination are not enrolled. Because eligibility is based on clinical suspicion or diagnosis of ocular motility abnormality rather than confirmed disease at enrollment, some participants may ultimately be clinically judged to have no abnormality on formal evaluation; such subjects remain part of the study population. Participants will undergo a single study visit consisting of (1) a standard clinical 9-gaze examination with clinician grading (-4 to +4), (2) a smartphone video recording performed with the assistance of a trained operator using a standardized setup, and (3) a patient self-recorded smartphone video acquired through the study mobile application.

For quantitative reference measurements, independent rater-derived measurements will be obtained from the recorded videos using a calibrated reference approach. A calibration marker of known size will be placed on the glabella during video acquisition to enable real-world scale estimation for the reference measurements. The calibration marker is used only for rater-based reference measurement and is not used by the AI algorithm.

The primary objective is to evaluate agreement between AI-derived quantitative eye movement metrics and rater-derived measurements of corneal center displacement, expressed as distance (mm) and angular deviation. Secondary objectives are to evaluate correlation between AI-derived metrics and clinician-assessed 9-gaze grading, and to evaluate whether AI-derived metrics obtained from patient self-recorded videos are comparable to those obtained from assistant-aided recordings. The investigational device is used solely for video acquisition and analysis and does not influence clinical decision-making during the study.

• Glandy EOM (AI-based quantitative eye movement analysis software) (Device): Glandy EOM is an investigational software-only medical device developed by THYROSCOPE INC. It has not yet received regulatory approval and is used for research purposes only in this study. Glandy EOM processes smartphone-recorded 9-gaze facial videos to generate AI-derived quantitative eye movement metrics: * Corneal center displacement (mm) * Angular deviation (degrees) Video acquisition in this study includes (a) an assistant-aided recording using a standardized smartphone setup in the clinical environment and (b) a patient self-recorded video acquired through the study mobile application. A calibration marker of known size placed on the glabella is used for rater-derived reference measurements only and is NOT used by the AI algorithm. AI outputs are generated for research analysis and are NOT used to guide real-time clinical decisions during the study.

Primary outcomes

• Agreement between AI-derived and rater-derived corneal center displacement (distance, mm) [At study visit (single visit)]: Agreement between AI-derived and rater-derived measurements of corneal center displacement expressed as distance (mm) across 9 gaze positions, assessed by intraclass correlation coefficient (ICC) and Bland-Altman analysis.
• Agreement between AI-derived and rater-derived angular deviation (degrees) [At study visit (single visit)]: Agreement between AI-derived and rater-derived measurements of corneal center displacement expressed as angular deviation (degrees) across 9 gaze positions, assessed by ICC and Bland-Altman analysis.
• Linear association and absolute error vs. rater-derived reference [At study visit (single visit)]: Linear association and absolute error between AI-derived and rater-derived measurements (both distance and angular deviation), assessed by Pearson correlation coefficient and mean absolute error (MAE), respectively.

Secondary outcomes

• Correlation between AI-derived metrics (assistant-aided recording) and clinician-assessed 9-gaze grading [At study visit (single visit)]: Correlation between AI-derived eye movement metrics obtained from assistant-aided smartphone recordings and clinician-assessed 9-gaze grading (-4 to +4), assessed using Spearman correlation.
• Correlation between AI-derived metrics (patient self-recorded) and clinician-assessed 9-gaze grading [At study visit (single visit)]: Correlation between AI-derived eye movement metrics obtained from patient self-recorded smartphone videos and clinician-assessed 9-gaze grading (-4 to +4), assessed using Spearman correlation.
• Agreement between AI-derived metrics from patient self-recorded vs. assistant-aided recordings [At study visit (single visit)]: Agreement between AI-derived metrics obtained from patient self-recorded and assistant-aided smartphone videos, assessed using ICC and Bland-Altman analysis.
• Linear association and absolute error: patient self-recorded vs. assistant-aided [At study visit (single visit)]: Linear association and absolute error between AI-derived metrics from patient self-recorded and assistant-aided smartphone videos, assessed using Pearson correlation coefficient and mean absolute error (MAE).

Sex: All, ages: 22 Years to any

Inclusion Criteria:

* Adults who have reached the legal age of consent for research participation in the applicable jurisdiction.
* Subjects with suspected or known ocular motility abnormalities.
* Subjects who are able to understand and follow gaze instructions required for the 9-gaze examination, as judged by the investigator.
* Subjects who are able to undergo facial video recording using a smartphone.
* Subjects who provide written informed consent.

Exclusion Criteria:

* Conditions that prevent reliable detection of ocular landmarks in video recordings (e.g., severe ptosis, corneal opacity, or significant occlusion of the eyes).
* Ocular or facial conditions that, in the opinion of the investigator, may significantly interfere with image-based analysis.
* Subjects who are unable to comply with study procedures.