MedWGAN based synthetic dataset generation for Uveitis pathology

Clinical decision support based on artificial intelligence (AI) methods has increasingly been employed in medical applications to diagnosis. Developing efficient AI methods, however, depends necessarily the availability of sufficiently large amount data provide reliable results. But, medicine, it is not always possible find sufficient real all pathologies, particularly, for rare diseases. This paper proposes a methodological framework generating synthetic using augmentation techniques combined with epidemiological profiles. It focuses Uveitis, disease ophthalmology, which difficult diagnose because disparity prevalence its etiologies. The generated have qualitatively validated by specialist ophthalmologists and quantitatively tested machine learning methods. Results show that, randomly selected sample data, more than 55% were assessed as good or excellent, very promising synthetic, near-real, They also that proposed consistent Uveitis pathology, different dataset sizes, achieving 80% diagnosis prediction accuracy 2000 patient records larger.