Using an Ensemble of GANs and CNNs to More Accurately Generate and Diagnose Skin Condition Datasets in Diverse Skin Types
Generated and validated synthetic images to address ethical AI bias due to lack of diverse skin condition images.
Skin Ensemble
Skin Ensemble uses an ensemble of GANs and CNNs to generate and diagnose skin condition datasets, focusing on diversity and ethical AI.
Motivation
Medical AI models often lack training data for diverse skin types, leading to bias. This project generates synthetic images to fill those gaps and validates them with CNNs.
Approach
- GANs generate realistic skin condition images for underrepresented skin types
- CNNs validate the generated images for accuracy
- Ensemble methods improve robustness
Results
- Improved dataset diversity
- Better diagnostic accuracy across skin types
Built with PyTorch, TensorFlow, and medical datasets.