Che Azemin, Mohd Zulfaezal and Mohd Tamrin, Mohd Izzuddin and Hilmi, Mohd Radzi and Mohd. Kamal, Khairidzan (2023) Synthetic pterygium images using Style Generative Adversarial Networks (SGANs). In: 4th Optometry Scientific Conference, 12-13 August 2023, Bangi, Selangor. (Unpublished)
PDF
- Supplemental Material
Restricted to Registered users only Download (1MB) | Request a copy |
|
PDF (Tentative Program)
- Supplemental Material
Restricted to Repository staff only Download (350kB) | Request a copy |
Abstract
PURPOSE: Pterygium, an ocular surface disorder, poses diagnostic challenges for optometrist and ophthalmologist. We proposed using Style-Generative Adversarial Networks (SGANs) to generate synthetic pterygium images for training purposes. METHODS: A training datasets of 68 pterygium images collected during routine clinical examinations was used. Fréchet inception distance (FID) was employed to evaluate the similarity between the synthetic and original images. RESULTS: FID analysis revealed that the synthetic images closely resemble the original pterygium images, suggesting a high degree of similarity. This indicates the potential of SGANs in generating realistic pterygium images. CONCLUSIONS: The successful generation of synthetic pterygium images using SGANs provides a valuable tool for training future optometrist and ophthalmologist in pterygium diagnosis and grading. By expanding the availability of diverse pterygium images, trainees can enhance their skills and proficiency. The use of synthetic images overcomes limitations associated with obtaining a sufficient number of real pterygium images. Additionally, the availability of a large dataset of synthetic images enables the development of advanced machine learning algorithms and computer-assisted diagnostic tools, improving the accuracy and efficiency of pterygium grading. SGAN generated images have the potential to standardize and control the training process, leading to improved patient care and management of pterygium.
Actions (login required)
View Item |