Skip to content
DraconiticResearch
Search Sources API
201,818 studies31 sourcesUpdated 22 min ago
Peer-reviewedCC BY-NC-NDPhotoacoustics

High-fidelity dual-domain photoacoustic data synthesis with latent diffusion models

  • Linxin Li1
  • Zhiyuan Zheng1
  • Ziyang Wang1
  • Ye Yu1
  • Haipeng Wei1
  • Shangkun Hou1
  • Hanchen Xu1
  • Qiegen Liu1
  • Xianlin Song2
  1. 1School of Information Engineering, Nanchang University, Nanchang, 330031, China
  2. 2Corresponding author.
Read full textPDFDOICite

Abstract

Photoacoustic tomography (PAT) is widely utilized in biomedical imaging due to its high optical contrast and deep acoustic penetration. However, the scarcity of high-quality datasets hinders the application of deep learning in this field. To address this, we propose a latent diffusion model (LDM)-based dual-domain data generation framework, which consists of two independently-trained synthesis pathways, that synthesizes photoacoustic data in both image and projection domains. Our method maps raw data into a downsampled latent space, preserving essential structural and semantic information while significantly reducing modeling dimensionality. The two pathways are trained separately to model their corresponding data distributions. To validate the framework, the generated data was used to train a reconstruction network for sparse-view, limited-view, and bandwidth-limited tasks. Experimental results show that models trained with synthetic data achieve performance comparable to reference data training in terms of structural fidelity, artifact suppression, and overall image quality. Quantitative evaluation via PSNR and SSIM metrics further confirms the high usability of the generated data. This study represents the first systematic exploration of dual-domain generation in PAT, offering a robust strategy for deep learning research under medical data constraints.

Related

  1. EPC-3D-Diff: Equivariant Physics Consistent Conditional 3D Latent Diffusion for CBCT to CT Synthesis
    Alzahra Altalib, Chunhui Li, Haytham Ahmad Alewaidat +3
    arXivPreprintCC BY 4.0Full textBiological & medical physicsarXiv
  2. Acquisition-inspired motion artifact modeling and degradation-consistent diffusion restoration in photoacoustic microscopy
    Heqing Wang, Shuyan Zhang, Jingtan Li +3
    PhotoacousticsPeer-reviewedCC BYFull textOptics & photonicsDOAJ
  3. Accelerating High‐Fidelity Airfoil Design via Physics‐Informed Video Diffusion Model
    Yifeng Ai, Yijia Guo, Hongfu Zhang +7
    Advanced SciencePeer-reviewedCC BY 4.0Full textStatistical & nonlinear physicsPMC
  4. Transport fidelity and domain of validity of compact Gaussian kinetic representations for rarefied flows
    Ehsan Roohi
    arXivPreprintCC BY 4.0Full textFluid dynamicsarXiv
  5. DAFNet: Dual-path Adaptive Fusion Network for High-Fidelity Cross-Modality Brain MR Image Synthesis
    Jialin Liu, Xiaoliang Zhang
    arXivPreprintarXiv licenceFull textBiological & medical physicsarXiv

Cite

BibTeXRIS
BibTeX
@article{li2026highfidelity,
  title = {{High-fidelity dual-domain photoacoustic data synthesis with latent diffusion models}},
  author = {Linxin Li and Zhiyuan Zheng and Ziyang Wang and Ye Yu and Haipeng Wei and Shangkun Hou and Hanchen Xu and Qiegen Liu and Xianlin Song},
  journal = {Photoacoustics},
  year = {2026},
  month = oct,
  doi = {10.1016/j.pacs.2026.100898},
  url = {https://doi.org/10.1016/j.pacs.2026.100898},
  note = {CC BY-NC-ND},
}
RIS
TY  - JOUR
TI  - High-fidelity dual-domain photoacoustic data synthesis with latent diffusion models
AU  - Linxin Li
AU  - Zhiyuan Zheng
AU  - Ziyang Wang
AU  - Ye Yu
AU  - Haipeng Wei
AU  - Shangkun Hou
AU  - Hanchen Xu
AU  - Qiegen Liu
AU  - Xianlin Song
PY  - 2026
DA  - 2026/10/19
JO  - Photoacoustics
DO  - 10.1016/j.pacs.2026.100898
UR  - https://doi.org/10.1016/j.pacs.2026.100898
AB  - Photoacoustic tomography (PAT) is widely utilized in biomedical imaging due to its high optical contrast and deep acoustic penetration. However, the scarcity of high-quality datasets hinders the application of deep learning in this field. To address this, we propose a latent diffusion model (LDM)-based dual-domain data generation framework, which consists of two independently-trained synthesis pathways, that synthesizes photoacoustic data in both image and projection domains. Our method maps raw data into a downsampled latent space, preserving essential structural and semantic information while significantly reducing modeling dimensionality. The two pathways are trained separately to model their corresponding data distributions. To validate the framework, the generated data was used to train a reconstruction network for sparse-view, limited-view, and bandwidth-limited tasks. Experimental results show that models trained with synthetic data achieve performance comparable to reference data training in terms of structural fidelity, artifact suppression, and overall image quality. Quantitative evaluation via PSNR and SSIM metrics further confirms the high usability of the generated data. This study represents the first systematic exploration of dual-domain generation in PAT, offering a robust strategy for deep learning research under medical data constraints.
SN  - 2213-5979
ER  -