Artificial Intelligence Diagnosis of Ocular Motility Disorders From Clinical Videos
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
BACKGROUND: Multimodal artificial intelligence (AI) models have recently expanded into video analysis. In ophthalmology, one exploratory application is the automated detection of extraocular movement (EOM) disorders. This proof-of-concept study evaluated the feasibility of using Gemini 2.0 to recognize EOM abnormalities, identify the affected eye, and recognize specific movement limitations from publicly available, real-world clinical videos. METHODS: We retrospectively collected 114 YouTube videos of EOM disorders, including cranial nerve (CN) palsies, internuclear ophthalmoplegia (INO), supranuclear disorders, nystagmus, and ocular myasthenia gravis (MG), alongside 15 control videos demonstrating normal EOMs. Videos were trimmed to include only the pertinent clinical examinations, and audio was removed to avoid diagnostic cues. Using a standardized zero-shot prompt, Gemini 2.0 analyzed each video via the Google AI Studio platform. Gemini 2.0 was evaluated based on its ability to provide the correct diagnosis, identify the affected eye, and recognize the specific movement limitation (if any). Descriptive statistics, Spearman correlations, and comparative analyses were used to assess performance. RESULTS: Gemini 2.0 correctly identified the primary diagnosis in 43 of 114 videos, yielding an overall diagnostic accuracy of 37.7%. Diagnostic performance varied by condition, with the highest accuracies observed in third nerve palsy (81.1%), INO (80.0%), sixth nerve palsy (66.7%), and ocular MG (20.0%), whereas normal EOMs were correctly classified in 93.3% of cases. In misclassified cases, the correct diagnosis appeared in the differential diagnosis in 15.5% of instances. Laterality was correctly identified in 26.5% of eligible cases overall, 73.1% among correctly diagnosed cases vs. 9.6% in misclassified ones. Similarly, movement limitations were accurately identified in 30.3% of eligible cases overall, with a marked increase to 88.5% accuracy in correctly diagnosed cases compared to 9.6% in misclassified cases. Longer videos moderately correlated with longer processing time (ρ = 0.55, P < 0.001). Significant correlations were observed between correct diagnosis and correct laterality identification (ρ = 0.45, P < 0.001), correct diagnosis and correct movement limitation identification (ρ = 0.61, P < 0.001), and laterality and movement limitation (ρ = 0.51, P < 0.001). Processing time averaged 11.0 seconds and correlated with video length (ρ = 0.55, P < 0.001). CONCLUSIONS: This proof-of-concept study demonstrates the feasibility of using Gemini 2.0 for automated recognition of EOM abnormalities in clinical videos. Although performance was stronger in overt cases, overall diagnostic accuracy remains limited. Substantial validation on standardized, clinician-annotated datasets is needed before clinical application.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.002 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it