Democratising Clinical AI through Dataset Condensation for Classical Clinical Models

Researchers have developed a new framework that enables dataset condensation for non-differentiable clinical AI models like decision trees and Cox regression, using differential privacy to create synthetic medical datasets. This breakthrough allows healthcare institutions to share condensed synthetic data while preserving patient privacy and maintaining model utility across classification and survival prediction tasks.

• →Dataset condensation traditionally required differentiable neural networks, limiting its use with popular clinical models like decision trees.
• →The new zero-order optimization framework extends dataset condensation to non-differentiable models using only function evaluations.
• →Differential privacy guarantees enable safe sharing of synthetic clinical data without exposing sensitive patient information.
• →Empirical testing across six datasets demonstrates preserved model utility for both classification and survival prediction tasks.
• →This advancement could democratize access to clinical AI training data while maintaining strict privacy protections.