DiGAN Breakthrough: Advancing Diabetic Data Analysis with Innovative GAN-Based Imbalance Correction Techniques

📝 Abstract

DiGAN represents a breakthrough approach in the field of medical diagnostics, especially in diabetes classification. Clinical diabetic datasets are frequently characterized by severe class imbalance, where minority classes represent critical high-risk patient groups. Standard oversampling techniques often fail to generate sufficiently realistic synthetic samples for complex clinical feature spaces. DiGAN employs a Generative Adversarial Network architecture specifically designed for diabetic data distributions, incorporating domain knowledge into the discriminator to ensure physiological plausibility of generated samples. The augmented datasets consistently improve classifier performance, achieving significant gains in minority class recall and AUC compared to existing baselines.

📋 BibTeX Citation

@article{zhao2024digan,
  title     = {{DiGAN} Breakthrough: Advancing Diabetic Data 
               Analysis with Innovative {GAN}-Based Imbalance 
               Correction Techniques},
  author    = {Zhao, Puyang and Liu, X. and Yue, Z. and others},
  journal   = {Computer Methods and Programs in 
               Biomedicine Update},
  year      = {2024},
  month     = {apr},
  doi       = {10.1016/j.cmpbup.2024.100152},
  url       = {https://doi.org/10.1016/j.cmpbup.2024.100152},
  publisher = {Elsevier}
}

🔗 Related Publications