Diagnosing lagophthalmos using artificial intelligence.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Lagophthalmos, incomplete eyelid closure, risks corneal ulceration and blindness.
• It is a symptom associated with various underlying medical conditions.
• Current diagnostic methods may not be rapid or consistently reliable.

Purpose of the Study:

• To develop and validate a convolutional neural network (CNN) for automated lagophthalmos diagnosis.
• To assess the accuracy and efficiency of the CNN model in identifying lagophthalmos.
• To explore the potential of AI in improving clinical workflows for lagophthalmos patient care.

Main Methods:

• Prospective data collection from 30 lagophthalmos patients and 10 healthy controls.
• Training a CNN model using 826 images, with validation and testing sets of 91 images each.
• Utilizing anti-overfitting strategies during model training.

Main Results:

• The CNN model achieved a validation accuracy of 97.8% and a testing accuracy of 93.41%.
• The model training was completed in 17.3 minutes.
• Low average and final loss values were recorded during training and validation.

Conclusions:

• The developed CNN offers a novel, rapid, and reliable method for lagophthalmos diagnosis.
• The AI-based approach demonstrates high accuracy and efficiency.
• This tool has significant potential to enhance physician workflow and improve patient care for lagophthalmos.