Pith. sign in

REVIEW 3 major objections 4 minor 59 references

A Conditional Diffusion Model for Electrical Impedance Tomography Image Reconstruction

T0 review · 3 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Conditional diffusion model reconstructs EIT images from boundary voltages and beats all compared methods on simulated tests.

desk verdict A solid, modest application of conditional diffusion to EIT with a real SOTA bump on clean simulated data, but the sim-to-real normalization is a calibration-based heuristic, not a derivation, and the abstract overclaims. read the letter →

arxiv 2412.16979 v1 pith:HFI5PKE7 submitted 2024-12-22 cs.CV

Shuaikai Shi , Ruiyuan Kang , Panos Liatsis This is my paper · ORCID
classification cs.CV
keywords electricalimpedancetomographyimagereconstructionconditionaldiffusionmodeldenoisingprobabilisticTransformerU-netDDIMsamplingill-posedinverseproblemdomaingeneralization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proposes CDEIT, a generative model that reconstructs cross-sectional conductivity images from electrical impedance tomography (EIT) boundary-voltage measurements. Instead of regressing an image from voltage data in one step, CDEIT adds Gaussian noise to a clean conductivity image and then learns to denoise it back to a clean image while conditioning on the measured boundary voltages at every step. The authors argue that this multi-step, voltage-conditioned denoising implicitly learns a prior over conductivity images and refines spatial details progressively, which is why it reports higher accuracy than single-step deep-learning and classical methods on a simulated test set. The paper also claims that a voltage-and-current normalization procedure lets a model trained on simulated data be applied to real water-tank datasets with different sizes, excitation currents, and background conductivities without retraining.

What carries the argument

The central objects are the forward diffusion chain $\sigma_t = \sqrt{\bar{\alpha}_t}\,\sigma_0 + \sqrt{1-\bar{\alpha}_t}\,\epsilon$, which turns a clean conductivity image into Gaussian noise over $T=1000$ steps; the conditional reverse network $\sigma_\theta(\sigma_t, U, t)$, a Transformer-based U-net with global and windowed Swin attention, time-step embedding, patch merging, and skip connections, which estimates the clean image from the noisy image and the upsampled boundary-voltage map; the DDIM sampler, which allows the reverse process to run in five steps; and the normalization rule $\delta\hat{\sigma} = g\left(\frac{U_{11}-U_{10}}{S_U}\right)\cdot \frac{S_I}{S_U}$, which rescales real-world voltage differences by factors derived from a simplified resistor model, an assumed linear current scaling, and a voltage ratio measured from the target dataset's reference voltages.

What would settle it

Withhold the reference voltage $U_{10}$ used to compute $S_U$ from a real dataset, derive $S_U$ instead from the forward model or from a separately calibrated phantom, and run the reconstructions; if accuracy collapses, the claimed parameter-free domain transfer is refuted. A second check is already in the paper: at 30 dB input noise, CDEIT's PSNR drops below DHU-Net and Ec-Net, so the superiority claim is restricted to clean or mildly noisy measurements.

Watch

Extended reading notes

Core claim

The central discovery is that a denoising diffusion probabilistic model, re-purposed as a conditional generator, can solve the EIT inverse problem by modeling the conditional distribution $p(\sigma|U)$ of conductivity images given boundary voltages. In training, Gaussian noise is added to clean conductivity images, and a Transformer-based U-net learns to predict the original image from the noisy image together with the voltage measurements and the diffusion time step. At inference, a DDIM sampler runs the learned reverse process in as few as five steps, starting from pure noise and the measured voltages, to generate the conductivity image. On the simulated benchmark the method reaches a PSNR of 39.57 dB, SSIM of 0.998, and correlation coefficient of 0.999, outperforming all eight compared methods, including another diffusion-based approach that is not conditioned on voltages. The paper further claims that the model transfers to real water-tank data through a normalization procedure that compensates for differences in excitation current and background conductivity.

Load-bearing premise

The load-bearing premise is that the normalization formula, with the current scale factor $S_I$ assumed linear and the voltage scale factor $S_U$ measured from the target dataset's reference voltages, correctly maps real voltage differences into the simulation's units; if the simplified resistor model $R \approx \frac{2}{\pi\sigma}\ln\frac{2d}{\delta}$ fails under electrode contact impedance and nonlinear current distribution, the reported real-data generalization is not parameter-free.

Editorial extensions

If this is right

  • On the simulated test set, CDEIT achieves PSNR 39.57 dB, SSIM 0.998, and CC 0.999, outperforming all compared conventional and deep-learning methods.
  • The DDIM sampler with five reverse steps keeps most of the quality gain while bringing inference to 1.55 GFLOPs, less than several lighter-parameter baselines.
  • The proposed normalization lets a model trained on simulated data produce usable reconstructions on two real water-tank datasets with different sizes, currents, and background conductivities, without retraining.
  • At low noise (40 dB), CDEIT keeps the top PSNR, but at 30 dB the single-step baselines DHU-Net and Ec-Net overtake it, bounding the reported advantage to clean or mildly noisy measurements.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The conditional-diffusion formulation naturally supports uncertainty quantification: sampling several reconstructions from the same voltage vector would yield per-pixel variance estimates, which the paper does not compute.
  • The normalization procedure compensates only for global scalings of current and conductivity; extending it to electrode contact impedance or non-circular domains would require a richer transfer model and would directly test the simplified resistor assumption.
  • The 30 dB noise result suggests that training the denoiser with voltage noise, or conditioning on an estimated noise level, could make generative refinement robust where single-step regression currently wins.
  • If $S_U$ could be estimated from the forward model instead of from the target dataset's own reference voltages, the simulation-to-real transfer would become truly parameter-free; as presented, the scale factor is calibrated on the test domain.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper proposes CDEIT, a conditional denoising diffusion model for electrical impedance tomography (EIT) image reconstruction. The model adds Gaussian noise to clean conductivity images in a forward diffusion process and trains a Transformer-based U-Net to reverse this process conditioned on boundary voltage measurements, using a standard DDPM-style loss with an L1 simplification. Inference is performed with a DDIM sampler using only five steps. The authors report state-of-the-art performance on a simulated dataset (PSNR 39.57, SSIM 0.998, CC 0.999) and propose a normalization procedure in Section IV intended to allow models trained on simulated data to be applied to real datasets with different sizes, excitation currents, and background conductivities. Experiments on two real datasets (UEF2017 and KTC2023) are presented, but with only four hand-picked samples and manually labeled references.

Significance. If the simulated-dataset results hold, the paper provides a useful and well-documented application of conditional diffusion models to EIT, with open-source code and a standard derivation of the DDPM objective. The reported 0.81 dB PSNR improvement over DHU-Net on the test set is a modest but consistent gain, and the ablation and complexity analyses are reasonable. However, the third contribution, sim-to-real generalization, is not established: the normalization derivation in Section IV is not physically valid as presented, and the real-data evaluation is too limited to support the claimed generalization. The abstract's blanket claim of outperforming state-of-the-art methods is also contradicted by the paper's own noise-robustness results in Table IV.

major comments (3)
  1. [Abstract, Table IV, Sec. V-E3] The abstract states that the proposed model 'outperforms state-of-the-art methods' without qualification, but Table IV shows that at 30 dB Gaussian noise the CDEIT PSNR is 26.70, below DHU-Net (28.94) and Ec-Net (28.78). The text in Sec. V-E3 explicitly acknowledges this degradation. The SOTA claim must be restricted to the noise-free or low-noise setting, or the claim should be removed from the abstract.
  2. [Sec. IV, Eqs. (26)-(32)] The normalization derivation is not a valid transfer law. Eq. (26) is a two-point resistance formula for an extended conductor and does not describe the response of a 16-electrode setup governed by the complete electrode model with contact impedances z_l in Eq. (2). Eqs. (27)-(28) write I = Uσ, but σ is a conductivity distribution, not a scalar conductance, and in EIT the measurement electrodes draw essentially no current; the product Uσ is not the current flowing between measurement electrodes. The key assertion SI = I10/I00 = I11/I01, attributed to 'linearity', is exactly the property that needs proof, since the forward map f(σ) is nonlinear and the domain mismatch may scale differently in the background and inclusion frames. Finally, SU is computed from reference voltages of the target real dataset (text following Eq. (32)), so the claimed 'prediction' on real data is calibrated on the deployment domain rather than derived from training-domain quantities; calling this 'directly applied... without retraining' overstates what was actually tested.
  3. [Sec. V-D, Fig. 10, Fig. 11] The real-data validation uses only four hand-picked samples with manually labeled reference images, and the text states that the quantitative indicators are 'shown for reference purposes only'. This does not provide a quantitative test of the normalization procedure; the sim-to-real generalization claim rests entirely on the derivation in Sec. IV, which has the problems noted above. A more systematic evaluation on the full real datasets, or at least a clearly stated limitation, is needed.
minor comments (4)
  1. [Sec. III-D, Eq. (20)] Eq. (20) retains a negative coefficient from the KL divergence in Eq. (18), but the simplified loss in Eq. (21) is a positive L1 norm. Since the L1 loss is what is actually optimized, this is a presentation inconsistency rather than a fatal error, but the sign should be clarified.
  2. [Sec. V-E1, Table II] There is a typo, 'bilion' instead of 'billion', in the complexity discussion.
  3. [Sec. VI] The conclusion contains a typo, 'generalizion' instead of 'generalization'.
  4. [References] The reference list has inconsistent author name spellings, e.g., 'Husain' in [14] versus 'Hussain' in [6] and [7], which should be unified.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: the DDPM/ELBO/loss chain is standard and self-contained, and the Sec. IV SU calibration is an explicit target-domain constant, not a conclusion-forcing fit.

full rationale

The conditional-diffusion derivation (Secs. III-B through III-F and the Appendix) is self-contained: the forward Gaussian process (Eq. 11), the ELBO decomposition (Eq. 37), the posterior mean (Eq. 16), and the simplified ℓ1 objective (Eq. 21) follow standard algebra and do not assume the reconstruction target. The simulated-data SOTA claim (Table I) is an independent benchmark comparison against external methods. The only candidate for circularity is the Sec. IV transfer formula: Eq. (32) is applied using SU ≈ Ur/Us, where Ur is the average reference voltage from the target real dataset. That is an explicit calibration constant, not a fitted quantity that already encodes the conductivity-change answer; the network g must still produce the image from a normalized voltage difference. The paper discloses this calibration and the absence of real ground truth. The unsupported 'due to linearity' assertion (SI = I10/I00 = I11/I01) and the use of the two-point resistance formula (Eq. 26) are physical/mathematical correctness concerns, not by-construction equivalences. The real-data evaluation is limited to four hand-picked samples with manually labeled references (Sec. V-D), which weakens the generalization claim but does not create circularity. Reference [25] is a non-load-bearing self-citation used only as an example of conditional diffusion image fusion. No prediction was found to reduce to its input by construction.

Assumptions & free parameters 6 free parameters · 5 assumptions · 1 invented entities

The central result depends on standard DDPM machinery (noise schedule, Gaussian posterior) and on the forward EIT model embodied in the simulation. The loader's main custom assumptions are the simplified resistor model and the linear current scaling in the normalization scheme, plus the calibration constant SU measured from target-domain reference data.

free parameters (6)
  • voltage scaling factor SU = Ur/Us (ratio of reference voltages)
    Used in Eq. (32) to normalize real voltages; computed from reference measurements from the target real dataset, so it is a calibration constant fitted to the target domain.
  • diffusion noise schedule {beta_t} = uniform from 1e-4 to 2e-2
    Chosen hyperparameters for the forward diffusion process, standard in DDPM.
  • training iterations = 150k
    Chosen to train the model; no criterion given.
  • DDIM sampling steps = 5
    Chosen for fast sampling; the paper says it delivers similar performance to multi-stage fusion models but no ablation on number of steps is shown.
  • background conductivity sigma00 = 1 S/m
    Simulated dataset background conductivity used as the reference for normalization; arbitrary choice.
  • inclusion conductivities = 0.01 and 2 S/m
    Chosen for insulating and conductive inclusions in the simulated dataset.
assumptions (5)
  • domain assumption 2D complete electrode model (CEM) equations (1)-(4) with 16 electrodes and adjacent excitation/measurement
    Physical forward model used to generate simulated data; assumes 2D domain, isotropic conductivity, known contact impedances.
  • standard math Gaussian posterior q(sigma_{t-1}|sigma_t, sigma_0) is Gaussian due to self-conjugacy
    Used in Eq. (15) to derive the KL loss; standard DDPM result.
  • domain assumption Resistance approximation R ≈ 2/(πσ) ln(2d/δ) and geometric scaling invariance
    Used in Sec. IV to justify size-invariance of voltages; ignores electrode contact impedance and 3D effects.
  • ad hoc to paper Linearity of the current scaling factor: SI = I10/I00 = I11/I01
    Asserted in Sec. IV to allow normalization across background and inclusion states; no justification provided.
  • standard math The reverse process p_theta(sigma_{t-1}|sigma_t) is Gaussian with fixed covariance gamma_t^2 I
    Standard DDPM approximation used in Eq. (12).
invented entities (1)
  • effective measurement currents I00, I01, I10, I11
    purpose: Linking voltage and conductivity via Ohm's law in the normalization derivation.
    The quantities I00, I01, I10, I11 are introduced in Eqs. (27)-(28) without a clear operational definition; they are not measurable and the derivation depends on their values.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Conditional Diffusion Model for Electrical Impedance Tomography Image Reconstruction." pith.science (2026). https://pith.science/paper/HFI5PKE7

@misc{pith2026241216979,
  author       = {Pith},
  title        = {Pith review of: A Conditional Diffusion Model for Electrical Impedance Tomography Image Reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HFI5PKE7}},
  note         = {Machine review of arXiv:2412.16979}
}
read the original abstract

Electrical impedance tomography (EIT) is a non-invasive imaging technique, capable of reconstructing images of the electrical conductivity of tissues and materials. It is popular in diverse application areas, from medical imaging to industrial process monitoring and tactile sensing, due to its low cost, real-time capabilities and non-ionizing nature. EIT visualizes the conductivity distribution within a body by measuring the boundary voltages, given a current injection. However, EIT image reconstruction is ill-posed due to the mismatch between the under-sampled voltage data and the high-resolution conductivity image. A variety of approaches, both conventional and deep learning-based, have been proposed, capitalizing on the use of spatial regularizers, and the paradigm of image regression. In this research, a novel method based on the conditional diffusion model for EIT reconstruction is proposed, termed CDEIT. Specifically, CDEIT consists of the forward diffusion process, which first gradually adds Gaussian noise to the clean conductivity images, and a reverse denoising process, which learns to predict the original conductivity image from its noisy version, conditioned on the boundary voltages. Following model training, CDEIT applies the conditional reverse process on test voltage data to generate the desired conductivities. Moreover, we provide the details of a normalization procedure, which demonstrates how EIT image reconstruction models trained on simulated datasets can be applied on real datasets with varying sizes, excitation currents and background conductivities. Experiments conducted on a synthetic dataset and two real datasets demonstrate that the proposed model outperforms state-of-the-art methods. The CDEIT software is available as open-source (https://github.com/shuaikaishi/CDEIT) for reproducibility purposes.

Figures

Figures reproduced from arXiv: 2412.16979 by the authors.

Figure 1
Figure 1. Overview of the proposed CDEIT framework, encompassing both the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 3
Figure 3. Framework of the denoising net σθ(σt, U, t). to derive the loss function. The evidence lower bound (ELBO) of the log-likelihood is given by [45]: L(θ) = − X T t=1 KL (q(σt−1|σt,σ0)||pθ(σt−1|σt)), (13) where KL(·||·) is the Kullback-Leibler divergence between the conditional posterior distribution and (12). A compre￾hensive derivation of (13) can be found in the Appendix. q(σt−1|σt,σ0) is the conditional posterior di… view at source ↗
Figure 5
Figure 5. Scale transformation modules. (a) Patch merging. (b) Upsampling [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figures from the paper (5 more)
Figure 6
Figure 6. Figure 6: Variations in size, excitation currents, and background conductivities. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Forward and inverse FEM mesh configurations. (a) Forward mesh. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 9
Figure 9. Figure 9: Reconstructed conductivity images from simulated test data samples. [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Reconstructed conductivity images from two real test data samples. [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: Reconstruction results for the various methods on 4 samples in real [PITH_FULL_IMAGE:figures/full_fig_p011_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 52 canonical work pages

  1. [43]

    A comparative study of variational autoencoders, normalizing flows, and score-based diffusion models for electrical impedance tomography,

    H. Wang, G. Xu, and Q. Zhou, “A comparative study of variational autoencoders, normalizing flows, and score-based diffusion models for electrical impedance tomography,” Journal of Inverse and Ill-posed Problems, vol. 32, no. 4, pp. 795–813, 2024

    work page 2024

  2. [44]

    Data-driven approaches for electrical impedance to- mography image segmentation from partial boundary data,

    A. Denker, ˇZeljko Kereta, I. Singh, T. Freudenberg, T. Kluth, P. Maass, and S. Arridge, “Data-driven approaches for electrical impedance to- mography image segmentation from partial boundary data,” pp. 119– 139, 2024

    work page 2024

  3. [1]

    Robust imaging using electrical impedance tomography: review of current tools,

    B. Brazey, Y . Haddab, and N. Zemiti, “Robust imaging using electrical impedance tomography: review of current tools,” Proc. R. Soc. A , vol. 478, no. 2258, 2022

    work page 2022

  4. [2]

    Electrical impedance tomography in acute respiratory distress syndrome,

    M. C. Bachmann, C. Morais, G. Bugedo, A. Bruhn, A. Morales, J. B. Borges, E. Costa, and J. Retamal, “Electrical impedance tomography in acute respiratory distress syndrome,” Critical Care , vol. 22, pp. 1–11, 2018

    work page 2018

  5. [3]

    Realistic forward and inverse model mesh generation for rapid three-dimensional thoracic electrical impedance imaging,

    A. Zifan, P. Liatsis, and H. Almarzouqi, “Realistic forward and inverse model mesh generation for rapid three-dimensional thoracic electrical impedance imaging,” Computers in Biology and Medicine , vol. 107, pp. 97–108, 2019

    work page 2019

  6. [4]

    Stimulation and measurement patterns versus prior information for fast 3d eit: a breast screening case study,

    P. Kantartzis, M. Abdi, and P. Liatsis, “Stimulation and measurement patterns versus prior information for fast 3d eit: a breast screening case study,” Signal Processing, vol. 93, no. 10, pp. 2838–2850, 2013

    work page 2013

  7. [5]

    Application of process tomography to multiphase flow measurement in industrial and biomedical fields: A review,

    J. Yao and M. Takei, “Application of process tomography to multiphase flow measurement in industrial and biomedical fields: A review,” IEEE Sensors J., vol. 17, no. 24, pp. 8196–8205, 2017

    work page 2017

  8. [6]

    Tactile sensing using ma- chine learning-driven electrical impedance tomography,

    Z. Hussain, N. A. Madjid, and P. Liatsis, “Tactile sensing using ma- chine learning-driven electrical impedance tomography,” IEEE Sensors Journal, vol. 21, no. 10, pp. 11 628–11 642, 2021

    work page 2021

Show all 59 references
  1. [7]

    Improved tactile stimulus reconstruction in electrical impedance tomography using the discrete cosine transform and machine learning,

    Z. Hussain and P. Liatsis, “Improved tactile stimulus reconstruction in electrical impedance tomography using the discrete cosine transform and machine learning,” IEEE Sensors Journal , vol. 24, no. 14, pp. 22 884– 22 095, 2024

    2024

  2. [8]

    Resolution enhancement of electrical resistivity tomography based on deep learning,

    X. Liu, Q. Lu, and S. Liu, “Resolution enhancement of electrical resistivity tomography based on deep learning,” IEEE Geosci. Remote Sens. Lett. , vol. 20, pp. 1–5, 2023

    2023

  3. [9]

    Mukhopadhyay, Ed., Next generation sensors and systems , ser

    S. Mukhopadhyay, Ed., Next generation sensors and systems , ser. Smart Sensors, Measurement and Instrumentation. United States: Springer, Springer Nature, 2016

    2016

  4. [10]

    Enhanced circular dichroism for achiral sensing based on a DNA-origami-empowered anapole metasurface,

    C. Li, T. He, X. Yang, C. Feng, Z. Zhang, J. Zhu, S. Dong, Y . Shi, Z. Wei, H. Jiao, Y . Zhang, H. Liu, Z. Wang, and X. Cheng, “Enhanced circular dichroism for achiral sensing based on a DNA-origami-empowered anapole metasurface,” Nano Letters , vol. 24, no. 31, pp. 9451–9458,...

    2024

  5. [11]

    Scattering exceptional point in the visible,

    T. He, Z. Zhang, J. Zhu, Y . Shi, Z. Li, Z. Wei, Y . Li, Z. Wang, C.-W. Qiu, and X. Cheng, “Scattering exceptional point in the visible,” Light: Science & Applications, vol. 12, 09 2023

    2023

  6. [12]

    Advances in electrical impedance tomography inverse problem solution methods: From traditional regularization to deep learning,

    C. Dimas, V . Alimisis, N. Uzunoglu, and P. P. Sotiriadis, “Advances in electrical impedance tomography inverse problem solution methods: From traditional regularization to deep learning,” IEEE Access , vol. 12, pp. 47 797–47 829, 2024

    2024

  7. [13]

    Adler and D

    A. Adler and D. Holder, Electrical impedance tomography: methods, history and applications . CRC Press, 2021

    2021

  8. [14]

    Tactile sensing using machine learning-driven electrical impedance tomography,

    Z. Husain, N. A. Madjid, and P. Liatsis, “Tactile sensing using machine learning-driven electrical impedance tomography,” IEEE Sensors J. , vol. 21, no. 10, pp. 11 628–11 642, 2021

    2021

  9. [15]

    Review on electrical impedance tomog- raphy: Artificial intelligence methods and its applications,

    T. A. Khan and S. H. Ling, “Review on electrical impedance tomog- raphy: Artificial intelligence methods and its applications,” Algorithms, vol. 12, no. 5, 2019

    2019

  10. [16]

    Applied potential tomography,

    D. C. Barber and B. H. Brown, “Applied potential tomography,” J. Phys. E, Sci. Instrum. , vol. 17, no. 9, p. 723, sep 1984

    1984

  11. [17]

    Tikhonov regularization and total least squares,

    G. H. Golub, P. C. Hansen, and D. P. O’Leary, “Tikhonov regularization and total least squares,” SIAM J. Matrix Anal. Appl. , vol. 21, no. 1, pp. 185–194, 1999

    1999

  12. [18]

    Iterative reconstruction methods using regularization and optimal current patterns in electrical impedance tomography,

    P. Hua, E. Woo, J. Webster, and W. Tompkins, “Iterative reconstruction methods using regularization and optimal current patterns in electrical impedance tomography,” IEEE Trans. Med. Imag. , vol. 10, no. 4, pp. 621–628, 1991

    1991

  13. [19]

    Photo-realistic single image super-resolution using a generative adversarial network,

    C. Ledig, L. Theis, F. Husz ´ar, J. Caballero, A. Cunningham, A. Acosta, A. Aitken, A. Tejani, J. Totz, Z. Wang et al. , “Photo-realistic single image super-resolution using a generative adversarial network,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , 2017...

    2017

  14. [20]

    Deep unsupervised learning using nonequilibrium thermodynamics,

    J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, and S. Ganguli, “Deep unsupervised learning using nonequilibrium thermodynamics,” in Proc. Int. Conf. Learn. Represent. (ICLR) , 2015, pp. 2256–2265

    2015

  15. [21]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” in Adv. Neural Infor . Process. Syst. (NeurIPS), vol. 33, 2020, pp. 6840– 6851

    2020

  16. [22]

    Diffusion models beat GANs on image synthesis,

    P. Dhariwal and A. Nichol, “Diffusion models beat GANs on image synthesis,” in Proc. Annu. Conf. Neural Inf. Process. Syst. , vol. 34, 2021, pp. 8780–8794

    2021

  17. [23]

    Wavegrad: Estimating gradients for waveform generation,

    N. Chen, Y . Zhang, H. Zen, R. J. Weiss, M. Norouzi, and W. Chan, “Wavegrad: Estimating gradients for waveform generation,” in Proc. Int. Conf. Learn. Represent. (ICLR) , 2020

    2020

  18. [24]

    Image super-resolution via iterative refinement,

    C. Saharia, J. Ho, W. Chan, T. Salimans, D. J. Fleet, and M. Norouzi, “Image super-resolution via iterative refinement,” IEEE Trans. Pattern Anal. Mach. Intell. , pp. 1–14, 2022

    2022

  19. [25]

    Hyperspectral and multispectral image fusion using the conditional denoising diffusion probabilistic model,

    S. Shi, L. Zhang, and J. Chen, “Hyperspectral and multispectral image fusion using the conditional denoising diffusion probabilistic model,” arXiv preprint arXiv:2307.03423 , 2023

    2023 arXiv

  20. [26]

    High-resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR) , 2022

    2022

  21. [27]

    Palette: Image-to-image diffusion models,

    C. Saharia, W. Chan, H. Chang, C. Lee, J. Ho, T. Salimans, D. Fleet, and M. Norouzi, “Palette: Image-to-image diffusion models,” in ACM SIGGRAPH 2022 Conference Proceedings , 2022, pp. 1–10

    2022

  22. [28]

    Scalable diffusion models with transformers,

    W. Peebles and S. Xie, “Scalable diffusion models with transformers,” in Proc. IEEE Int. Conf. Comput. Vis. (ICCV) , 2023, pp. 4172–4182

    2023

  23. [29]

    Image reconstruction in electrical impedance tomography based on structure-aware sparse bayesian learning,

    S. Liu, J. Jia, Y . D. Zhang, and Y . Yang, “Image reconstruction in electrical impedance tomography based on structure-aware sparse bayesian learning,” IEEE Trans. Med. Imag. , vol. 37, no. 9, pp. 2090– 2102, 2018

    2018

  24. [30]

    Total variation regularization in electrical impedance tomography,

    A. Borsic, B. M. Graham, A. Adler, and W. R. Lionheart, “Total variation regularization in electrical impedance tomography,” 2007

    2007

  25. [31]

    Bayesian image reconstruction using weighted laplace prior for lung respiratory monitoring with electrical impedance tomography,

    Y . Wu, B. Chen, K. Liu, S. Huang, Y . Li, J. Jia, and J. Yao, “Bayesian image reconstruction using weighted laplace prior for lung respiratory monitoring with electrical impedance tomography,” IEEE Trans. In- strum. Meas. , vol. 72, pp. 1–11, 2023

    2023

  26. [32]

    Noser: An algorithm for solving the inverse conductivity problem,

    M. Cheney, D. Isaacson, J. Newell, S. Simske, and J. Goble, “Noser: An algorithm for solving the inverse conductivity problem,” Int. J. Imag. Syst. Technol.,, vol. 2, no. 2, p. 66—75, 1990

    1990

  27. [33]

    Reconstructions of chest phantoms by the d-bar method for electrical impedance tomogra- phy,

    D. Isaacson, J. Mueller, J. Newell, and S. Siltanen, “Reconstructions of chest phantoms by the d-bar method for electrical impedance tomogra- phy,” IEEE Trans. Med. Imag. , vol. 23, no. 7, pp. 821–828, 2004

    2004

  28. [34]

    Image reconstruction based on convolutional neural network for electrical resistance tomography,

    C. Tan, S. Lv, F. Dong, and M. Takei, “Image reconstruction based on convolutional neural network for electrical resistance tomography,” IEEE Sensors J. , vol. 19, no. 1, pp. 196–204, 2019

    2019

  29. [35]

    Image reconstruction for electrical impedance tomography based on spatial invariant feature maps and convolutional neural network,

    D. Hu, K. Lu, and Y . Yang, “Image reconstruction for electrical impedance tomography based on spatial invariant feature maps and convolutional neural network,” in 2019 IEEE International Conference on Imaging Systems and Techniques (IST) , 2019, pp. 1–6

    2019

  30. [36]

    Deep learning based cell imaging with electrical impedance tomography,

    Z. Chen, Y . Yang, J. Jia, and P. Bagnaninchi, “Deep learning based cell imaging with electrical impedance tomography,” in 2020 IEEE International Instrumentation and Measurement Technology Conference (I2MTC), 2020, pp. 1–6

    2020

  31. [37]

    Structure-aware dual-branch network for elec- trical impedance tomography in cell culture imaging,

    Z. Chen and Y . Yang, “Structure-aware dual-branch network for elec- trical impedance tomography in cell culture imaging,” IEEE Trans. Instrum. Meas. , vol. 70, pp. 1–9, 2021. 14 SUBMISSION TO IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. XX, NO. XX, 2024

    2021

  32. [38]

    High-resolution conductivity reconstruction by electrical impedance tomography using structure- aware hybrid-fusion learning,

    H. Yu, H. Liu, Z. Liu, Z. Wang, and J. Jia, “High-resolution conductivity reconstruction by electrical impedance tomography using structure- aware hybrid-fusion learning,” Computer Methods and Programs in Biomedicine, vol. 243, p. 107861, 2024

    2024

  33. [39]

    Shape reconstruction with multiphase conductivity for electrical impedance tomography using improved convolutional neural network method,

    Y . Wu, B. Chen, K. Liu, C. Zhu, H. Pan, J. Jia, H. Wu, and J. Yao, “Shape reconstruction with multiphase conductivity for electrical impedance tomography using improved convolutional neural network method,” IEEE Sensors J. , vol. 21, no. 7, pp. 9277–9287, 2021

    2021

  34. [40]

    Error-constraint deep learning scheme for electrical impedance tomography (eit),

    Q. Wang, H. Zhang, X. Li, X. Duan, J. Wang, R. Zhang, H. Zhang, Y . Ma, H. Wang, and J. Jia, “Error-constraint deep learning scheme for electrical impedance tomography (eit),” IEEE Trans. Instrum. Meas. , vol. 71, pp. 1–11, 2022

    2022

  35. [41]

    Electrical impedance tomography deep imaging with dual-branch u-net based on deformable convolution and hyper-convolution,

    Z. Wang, X. Li, Y . Sun, and Q. Wang, “Electrical impedance tomography deep imaging with dual-branch u-net based on deformable convolution and hyper-convolution,” IEEE Trans. Instrum. Meas. , vol. 73, pp. 1–16, 2024

    2024

  36. [42]

    Deepeit: Deep image prior enabled electrical impedance tomography,

    D. Liu, J. Wang, Q. Shan, D. Smyl, J. Deng, and J. Du, “Deepeit: Deep image prior enabled electrical impedance tomography,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 45, no. 8, pp. 9627–9638, 2023

    2023

  37. [45]

    Bounding evidence and estimating log-likelihood in vae,

    L. Struski, M. Mazur, P. Batorski, P. Spurek, and J. Tabor, “Bounding evidence and estimating log-likelihood in vae,” in Proc. 26th Int. Conf. Artif. Intel. Stat. (AISTATS) , 2023

    2023

  38. [46]

    Bishop, Pattern Recognition and Machine Learning

    C. Bishop, Pattern Recognition and Machine Learning . Springer, January 2006

    2006

  39. [47]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” in Proc. Annu. Conf. Neural Inf. Process. Syst. , vol. 30, 2017

    2017

  40. [48]

    Swin transformer: Hierarchical vision transformer using shifted windows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” in Proc. IEEE Int. Conf. Comput. Vis. (ICCV) , 2021

    2021

  41. [49]

    Searching for activation functions,

    P. Ramachandran, B. Zoph, and Q. V . Le, “Searching for activation functions,” arXiv preprint arXiv:1710.05941 , 2017

    2017 arXiv

  42. [50]

    Root Mean Square Layer Normalization,

    B. Zhang and R. Sennrich, “Root Mean Square Layer Normalization,” in Proc. Annu. Conf. Neural Inf. Process. Syst. , Vancouver, Canada, 2019

    2019

  43. [51]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980 , 2014

    2014 arXiv

  44. [52]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” in Proc. Int. Conf. Learn. Represent. (ICLR) , 2021

    2021

  45. [53]

    Jeans, Mathematical Theory of Electricity and Magnetism , ser

    J. Jeans, Mathematical Theory of Electricity and Magnetism , ser. Cam- bridge Library Collection - Physical Sciences. Cambridge University Press, 2009

    2009

  46. [54]

    Eidors: Towards a community-based extensible software base for eit,

    A. Adler and W. R. Lionheart, “Eidors: Towards a community-based extensible software base for eit,” in Proc. 6th Conf. Biomed. Appl. , 2005, pp. 1–4

    2005

  47. [55]

    The sheffield data collection system,

    B. H. Brown and A. D. Seagar, “The sheffield data collection system,” Clinical Physics and Physiological Measurement , vol. 8, no. 4A, p. 91, nov 1987

    1987

  48. [56]

    Open 2d electrical impedance tomogra- phy data archive,

    A. Hauptmann, V . Kolehmainen, N. M. Mach, T. Savolainen, A. Sepp ¨anen, and S. Siltanen, “Open 2d electrical impedance tomogra- phy data archive,” arXiv preprint arXiv:1704.01178 , 2017

    2017 arXiv

  49. [57]

    Kuopio tomography challenge 2023 – electrical impedance tomography competition and open dataset,

    M. R ¨as¨anen, P. Kuusela, J. Jauhiainen, M. Arif, K. Scheel, T. Savolainen, and A. Sepp ¨anen, “Kuopio tomography challenge 2023 – electrical impedance tomography competition and open dataset,” Applied Mathe- matics for Modern Challenges , vol. 2, no. 2, pp. 93–118, 2024

    2023

  50. [58]

    The ace1 electrical impedance tomography system for thoracic imaging,

    M. M. Mellenthin, J. L. Mueller, E. D. L. B. de Camargo, F. S. de Moura, T. B. R. Santos, R. G. Lima, S. J. Hamilton, P. A. Muller, and M. Alsaker, “The ace1 electrical impedance tomography system for thoracic imaging,” IEEE Trans. Instrum. Meas. , vol. 68, no. 9, pp. 3137–3150, 2019

    2019

  51. [59]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. Bovik, H. Sheikh, and E. Simoncelli, “Image quality assessment: from error visibility to structural similarity,” IEEE Trans. Image Process., vol. 13, no. 4, pp. 600–612, 2004

    2004

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.