DP-UOTM: A Differentially Private Unbalanced Optimal Transport Based Approach for High Quality Medical Image Synthesis

Jinnan He (), Ai Ran (), Fengchi Yuan () and Wai Kin Chan ()
Additional contact information
Jinnan He: Tsinghua University, Institute of Data and Information, Tsinghua Shenzhen International Graduate School
Ai Ran: Tsinghua University, Institute of Data and Information, Tsinghua Shenzhen International Graduate School
Fengchi Yuan: Tsinghua University, Institute of Data and Information, Tsinghua Shenzhen International Graduate School
Wai Kin Chan: Tsinghua University, Institute of Data and Information, Tsinghua Shenzhen International Graduate School

A chapter in AI, Society and Digital Transformation, 2026, pp 90-103 from Springer

Abstract: Abstract Deep learning has transformed AI-driven healthcare services through advanced medical imaging analysis, yet privacy concerns over protected health information persistently hinder data sharing in clinical service ecosystems. Conventional anonymization methods, while compliant with healthcare data governance standards, irreversibly degrade diagnostic utility through quality loss or incomplete de-identification. We propose DP-UOTM, a differentially private Unbalanced Optimal Transport-based framework for medical service systems, generating regulatory-compliant synthetic images that preserve diagnostic fidelity via optimal transport, enforce provable ( $$\epsilon $$ ϵ , $$\delta $$ δ )-DP guarantees against re-identification, and resolve class imbalances through label distribution calibration. Evaluated on benchmarks MNIST and Fashion-MNIST, DP-UOTM outperforms DP-GANs with $$\ge $$ ≥ 22% lower FID scores and $$\ge $$ ≥ 2% higher classification accuracy across privacy budgets. Clinical validation on chest X-ray services demonstrates synthetic data maintains $$\ge $$ ≥ 4% diagnostic accuracy improvement over raw data through class balancing. By enabling privacy-preserving substitutes for raw medical images, this framework directly supports federated healthcare services and multi-institutional collaborations while addressing critical needs in clinical AI service development—providing scalable, audit-ready synthetic datasets that comply with medical data regulations without compromising diagnostic utility, thereby bridging the gap between AI innovation and healthcare privacy requirements.

Keywords: private data synthesis; unbalanced optimal transport; differential privacy; medical decision making (search for similar items in EconPapers)
Date: 2026
References: Add references at CitEc
Citations:

There are no downloads for this item, see the EconPapers FAQ for hints about obtaining it.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:spr:lnopch:978-3-032-13116-4_8

Ordering information: This item can be ordered from
http://www.springer.com/9783032131164

DOI: 10.1007/978-3-032-13116-4_8

Access Statistics for this chapter

More chapters in Lecture Notes in Operations Research from Springer
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().