Privacy-preserving synthetic lung CT generation via geometric filtering
July 24, 2026
A new framework for synthetic CT slice generation uses Optimal Transport Conditional Flow Matching and a geometric filtering pipeline to balance realism and privacy. The approach achieved an FID of 0.3290 and a Privacy Preservation Score of 0.549, though it noted challenges in preventing patient-specific anatomical re-identification.
HOW THIS AFFECTS YOU
●
researcherYou can leverage this geometric filtering method to improve privacy-preserving GAN training.
●
healthThis highlights the ongoing difficulty in fully removing patient-specific anatomical identities from synthetic medical imaging.