An Outperforming Artificial Intelligence Model to Identify Referable Blepharoptosis for General Practitioners

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Blepharoptosis, or a droopy eyelid, requires accurate identification for timely referral and treatment.
• Current diagnostic methods may be limited in accuracy among non-specialists.
• There is a need for objective tools to aid in the diagnosis of blepharoptosis.

Purpose of the Study:

• To develop and evaluate an AI model for automatic detection of referable blepharoptosis.
• To compare the diagnostic performance of the AI model against non-ophthalmic physicians.

Main Methods:

• A convolutional neural network (CNN) was trained on 1000 retrospective eye images labeled by oculoplastic surgeons.
• The AI model performed binary classification to identify blepharoptosis (true and pseudoptosis) or a healthy eyelid.
• The same dataset was used to test the performance of three non-ophthalmic physicians.

Main Results:

• The CNN model achieved 92% sensitivity and 88% specificity.
• Non-ophthalmic physicians achieved a mean sensitivity of 72% and a mean specificity of 82.67%.
• The AI model demonstrated superior performance in identifying referable blepharoptosis compared to the physician group.

Conclusions:

• AI-powered tools can accurately identify blepharoptosis, outperforming non-specialist physicians.
• Artificial intelligence holds significant potential to assist general practitioners in diagnosing and referring blepharoptosis cases.
• AI can improve the efficiency and accuracy of blepharoptosis screening in primary care settings.