DISC: Self-Distilled LDM for Grading Prostate Cancer

Overview

Proposed a self-distillation strategy for Latent Diffusion Models to improve the accuracy in mask-to-histopathology image translation.

Key Results

• Augmented training with synthetic samples and achieved 97.35% AUC on rare-class cancer grading

Related Publication

DISC: Latent Diffusion Models with Self-Distillation from Separated Conditions for Prostate Cancer Grading - ISBI 2024, Extended version at SynData4CV @ CVPR 2024