Functional blepharoptosis screening with generative augmented deep learning from external ocular photography.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Functional blepharoptosis, a condition causing eyelid drooping, impacts vision and requires accurate diagnosis.
• Early detection and management of blepharoptosis are crucial for preventing visual impairment.

Purpose of the Study:

• To develop and validate a deep learning model for detecting functional blepharoptosis using external ocular photographs.
• To assess the performance enhancement of the model when trained with synthetic data generated by a StyleGAN model.

Main Methods:

• A dataset of 771 ocular photographs was curated, with 639 eyes diagnosed with functional blepharoptosis.
• A baseline deep learning model was trained and validated on a subset of the data.
• The training dataset was augmented with 2000 synthetic images generated by a StyleGAN model to train an enhanced model.

Main Results:

• The baseline model achieved a sensitivity of 0.68, specificity of 0.89, and AUC of 0.87.
• The GAN-augmented model demonstrated improved performance with a sensitivity of 0.95, specificity of 0.67, and AUC of 0.91.
• The augmented model showed a notable increase in sensitivity for detecting blepharoptosis.

Conclusions:

• Deep learning models can reliably detect functional blepharoptosis from standard eye photographs.
• The integration of synthetic data generated by generative adversarial networks (GANs) holds significant potential for enhancing the accuracy and robustness of diagnostic AI models in ophthalmology.