Deep Learning-Based Detection And Grading Of Pterygium Using Artificially Generated Data From Smartphone Images

This study aimed to develop and validate a deep learning model capable of detecting and grading pterygium tissue using smartphone images. The model was trained on a dataset consisting of 11,614 normal smartphone eye images and 23,250 artificially generated pterygium images derived from 274 original smartphone pterygium eye images. The proposed AI system was designed to enhance early detection, surgical time decision-making, and post-surgical recurrence monitoring.

This research was conducted at Vision Health Research Clinic, a specialized ophthalmology center focusing on anterior segment diseases and AI-assisted diagnostic tools. The study adhered to ethical guidelines of Semnan university of medical sciences.

A deep learning-based artificial intelligence model was developed and trained using a dataset including both real and artificially generated images. The artificially generated pterygium images were synthesized using advanced image augmentation, big generative adversarial network(BigGAN) which is generative modeling technique to enhance dataset diversity. The model's performance was validated against expert grading by two independent ophthalmologists and one labeling expert. Model accuracy, sensitivity, and specificity were assessed using a confusion matrix and receiver operating characteristic (ROC) analysis.

The deep learning model demonstrated an accuracy of 94.6% in detecting and grading pterygium tissue, as confirmed by ophthalmologists. The model successfully differentiated between normal and pterygium eyes and provided consistent grading aligned with clinical assessments. The confusion matrix analysis revealed a sensitivity of 92.3%, specificity of 95.1%, and a positive predictive value (PPV) of 93.7%. Additionally, the ROC analysis yielded an area under the curve (AUC) score of 0.96, indicating strong discriminatory performance. The use of artificially generated images significantly enhanced the model’s ability to operate that is train and test, across diverse cases.

This study highlights the potential of AI-driven diagnostic tools in ophthalmology, particularly for pterygium detection and grading. Automated screening can facilitate early detection, optimize surgical timing, and improve post-surgical recurrence monitoring. The integration of AI in ophthalmic care contributes to sustainable healthcare by reducing the burden on specialists and enhancing accessibility to timely diagnosis and management without commute.