Assessing the efficacy of StyleGAN 3 in generating realistic medical images with limited data availability
In this study, we leveraged StyleGAN 3 to synthesize high-fidelity images of pterygium, achieving significant strides in image realism as evidenced by low Fréchet Inception Distance (FID) scores. Our results demonstrate that StyleGAN 3 can intricately capture the textural nuances and vascular patter...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Proceeding Paper |
| Language: | English English English |
| Published: |
Association for Computing Machinery
2024
|
| Subjects: | |
| Online Access: | http://irep.iium.edu.my/112492/2/112492_Assessing%20the%20efficacy%20of%20StyleGAN%203.pdf http://irep.iium.edu.my/112492/3/112492_ICSCA%202024%2013th%20International%20Conference%20on%20Software%20and%20Computer%20Applications.pdf http://irep.iium.edu.my/112492/4/112492_Assessing%20the%20efficacy%20of%20StyleGAN%203_Scopus.pdf http://irep.iium.edu.my/112492/ https://dl.acm.org/doi/10.1145/3651781.3651810 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In this study, we leveraged StyleGAN 3 to synthesize high-fidelity images of pterygium, achieving significant strides in image realism as evidenced by low Fréchet Inception Distance (FID) scores. Our results demonstrate that StyleGAN 3 can intricately capture the textural nuances and vascular patterns distinctive to pterygium, with color tones and variations that closely mirror clinical photography. The generated images exhibit high equivariance to transformations, retaining their realism under various manipulations. Clinician reviews, expressed through confusion matrices, validated the authenticity of the synthetic images, although variations in individual assessments highlighted the challenges in differentiating between generated and real images. Ultimately, our findings confirm the efficacy of StyleGAN 3 in producing synthetic medical images that could potentially expand datasets for medical research and training, while also underscoring the necessity for diversity in training data and model tuning to achieve optimal realism. |
|---|