EUVIMEDEUVIMED entdecken
Crossref

Diff-ANO: Towards fast high-resolution ultrasound computed tomography via conditional consistency models and adjoint neural operators

Xiang Cao, Qiaoqiao Ding, Xinliang Liu, Xiaoqun Zhang, Lei Zhang

Inverse Problems · 10.11.2026

DOI: 10.1088/1361-6420/aec045 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Inverse Problems, online erschienen am 10.11.2026. Verfasst von Xiang Cao und 4 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Ultrasound computed tomography (USCT) reconstructs sound speed from incomplete, noisy scattered-wave measurements. Its instability arises from the nonlinear inverse map and acquisition geometry. Generative models can supply learned priors that stabilize such partial differential equation (PDE)-constrained inversions. In this work, we introduce Diff-ANO (Diffusion-based models with adjoint neural operators), which alternates conditional consistency model (CM) refinement with surrogate measurement-gradient optimization. One forward multigrid neural operator (MgNO) predicts the transmitter/receiver source-field bank; a conjugated residual combination of those fields then forms the adjoint action and sound-speed gradient. Across four acquisition geometries, our method delivers strong reconstruction quality with a 6.75×online speedup over solver-based refinement in a matched computational-efficiency comparison. Independent finer-grid WaveSim measurements reduce potential convergent Born series (CBS) inverse crime and yield reconstruction trends consistent with the matched-CBS study. Together with additional evaluations of intermediate refinements and ablations, these results support efficient data-informed, measurement-guided USCT reconstruction and motivate extension to more realistic acquisition physics and measured data.

Zitieren

APA

Cao, X., Ding, Q., Liu, X., Zhang, X., & Zhang, L. (2026). Diff-ANO: Towards fast high-resolution ultrasound computed tomography via conditional consistency models and adjoint neural operators. Inverse Problems. https://doi.org/10.1088/1361-6420/aec045

Vancouver

Cao X, Ding Q, Liu X, Zhang X, Zhang L. Diff-ANO: Towards fast high-resolution ultrasound computed tomography via conditional consistency models and adjoint neural operators. Inverse Problems. 2026. doi:10.1088/1361-6420/aec045

BibTeX
@article{Cao2026,
  author = {Cao, Xiang and Ding, Qiaoqiao and Liu, Xinliang and Zhang, Xiaoqun and Zhang, Lei},
  title = {Diff-ANO: Towards fast high-resolution ultrasound computed tomography via conditional consistency models and adjoint neural operators},
  journal = {Inverse Problems},
  year = {2026},
  doi = {10.1088/1361-6420/aec045}
}
Kostenlos starten →