F2FLDM: Latent Diffusion Models with Histopathology Pre-Trained Embeddings for Unpaired Frozen Section to FFPE Translation
Published in Winter Conference on Applications of Computer Vision (WACV), 2025
Developed LDMs with Parameter-Efficient Fine-Tuning (PEFT) and Histopathology Pre-Trained Embeddings for translating low-quality frozen section images to high-quality FFPE images, taking generative model hallucinations into account and improving classification performance from 81.99% to 94.64% AUCROC on low-quality frozen slides, while the performance on high-quality FFPE images reached 94.63%.
| Paper | Project Page |
Recommended citation: Man M. Ho, Shikha Dubey, Yosep Chong, Beatrice S. Knudsen, and Tolga Tasdizen. (2025). "F2FLDM: Latent Diffusion Models with Histopathology Pre-Trained Embeddings for Unpaired Frozen Section to FFPE Translation." WACV 2025.
Download Paper