Development and validation of a convolutional neural network to identify blepharoptosis.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Blepharoptosis causes reversible vision loss and can indicate serious neurological conditions.
• Current diagnosis relies on manual examination, and AI tools often require specialized settings.
• Existing AI for blepharoptosis typically focuses on eyelid position under ideal conditions.

Purpose of the Study:

• To develop and evaluate a deep learning model for detecting blepharoptosis using realistic periocular images.
• To assess the model's performance against human expert diagnosis in varied conditions.
• To explore the potential of AI in enhancing telemedicine for blepharoptosis screening.

Main Methods:

• A convolutional neural network (CNN) deep learning model was trained on high-quality periocular images.
• The dataset included images from patients with blepharoptosis and other eyelid conditions.
• Model performance was validated against nine medical experts and tested on a diverse image quality dataset.

Main Results:

• The deep learning model achieved an area under the receiver operating characteristic curve (AUC) of 0.918.
• The model's performance was statistically comparable to human expert graders, even with varied image quality.
• The AI demonstrated robust detection capabilities in realistic, non-specialized settings.

Conclusions:

• Deep learning models can effectively detect blepharoptosis in real-world conditions.
• AI-powered tools show significant potential for improving the accuracy and accessibility of telemedicine for eye conditions.
• This technology could aid in early detection and management of blepharoptosis, reducing vision loss.