EUVIMEDEUVIMED entdecken
Crossref

A Generative Model-Enhanced Deep Image Prior Framework for Unsupervised 3-D Electrical Impedance Tomography

Yuehui Wu, Xinhao Bai, Jiewei Lu, Jianda Han, Ningbo Yu

IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 1-15

DOI: 10.1109/tim.2026.3677975 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Yuehui Wu und 4 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Tianjin Municipality, China Postdoctoral Science Foundation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Electrical impedance tomography (EIT), a noninvasive, radiation-free, and real-time imaging modality of tissue conductivity changes, has become an important clinical tool. 3-D EIT offers superior spatial characterization of lung function, but its image reconstruction remains challenging due to the severely ill-posed inverse problem and high sensitivity to measurement noise. This article presents a generative model-enhanced deep image prior (DIP) framework for unsupervised 3-D EIT reconstruction, where the DIP provides implicit regularization and the generative model prior imposes explicit shape constraints to enforce a physiologically plausible image. A dual-output network is designed to simultaneously reconstruct the conductivity image and update the generator latent variable. The reconstruction is performed through instance-wise unsupervised optimization with a unified objective that combines data-consistency loss and shape constraint, allowing both the network parameters and the latent representation to be jointly updated. In addition, a residual Noise2Noise (R-N2N) denoiser module is devised for EIT data to suppress measurement noise in the voltage input by leveraging a residual learning strategy and reciprocal voltage pairs. Extensive simulations and experiments are conducted to evaluate the reconstruction quality, generalization performance, and denoising ability under diverse pulmonary ventilation conditions. The results demonstrate that the proposed framework significantly improves 3-D pulmonary ventilation imaging with accurate conductivity distribution, high shape fidelity, and effective artifact suppression.

Förderung

  • National Natural Science Foundation of China
  • Natural Science Foundation of Tianjin Municipality
  • China Postdoctoral Science Foundation

Zitieren

APA

Wu, Y., Bai, X., Lu, J., Han, J., & Yu, N. (2026). A Generative Model-Enhanced Deep Image Prior Framework for Unsupervised 3-D Electrical Impedance Tomography. IEEE Transactions on Instrumentation and Measurement, 75, 1-15. https://doi.org/10.1109/tim.2026.3677975

Vancouver

Wu Y, Bai X, Lu J, Han J, Yu N. A Generative Model-Enhanced Deep Image Prior Framework for Unsupervised 3-D Electrical Impedance Tomography. IEEE Transactions on Instrumentation and Measurement. 2026;75:1-15. doi:10.1109/tim.2026.3677975

BibTeX
@article{Wu2026,
  author = {Wu, Yuehui and Bai, Xinhao and Lu, Jiewei and Han, Jianda and Yu, Ningbo},
  title = {A Generative Model-Enhanced Deep Image Prior Framework for Unsupervised 3-D Electrical Impedance Tomography},
  journal = {IEEE Transactions on Instrumentation and Measurement},
  year = {2026},
  volume = {75},
  pages = {1--15},
  doi = {10.1109/tim.2026.3677975}
}
Kostenlos starten →