Pith. sign in

REVIEW 3 major objections 7 minor 40 references

QuantEIT: Ultra-Lightweight Quantum-Assisted Inference for Chest Electrical Impedance Tomography

T0 review · 3 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read QuantEIT claims that two parallel 2-qubit circuits plus a single linear layer reconstruct 2D and 3D chest EIT images as well as deep networks with roughly 500 times more parameters, and that this works without any training data.

desk verdict Plausible lightweight unsupervised EIT with a genuinely new quantum-circuit application, but the central 'quantum prior' claim rests on an ablation that its own equations refute. read the letter →

arxiv 2507.14031 v1 pith:JZ4M2OT2 submitted 2025-07-18 cs.CV cs.ETcs.LG

Hao Fang , Sihao Teng , Hao Yu , Siyi Yuan , Huaiwu He , Zhe Liu , Yunjie Yang This is my paper
classification cs.CVcs.ETcs.LG
keywords electricalimpedancetomographyquantumcircuitimagereconstructioninverseproblemunsupervisedlearningimplicitpriorlungimaginglightweightnetwork
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

The paper tries to establish that a tiny quantum circuit, not a deep neural network, can serve as the prior that tames an ill-posed inverse problem, and that for chest EIT this quantum prior is both smaller and more stable under noise than learned alternatives. It proposes QuantEIT, in which two parallel 2-qubit circuits produce a four-dimensional latent vector and a single linear layer turns that vector into a conductivity image; the whole system is optimized per measurement against the voltage data plus standard regularizers, with no training set. On 2D and 3D simulations, a lung phantom, and clinical data, the paper reports reconstruction accuracy comparable to or better than the Noser and R-SIP baselines while using only 0.2% of their parameters. If true, this would make hybrid quantum-classical models a practical route to lightweight, data-free reconstruction at the bedside and in other inverse problems.

What carries the argument

The QA-Net (Quantum-Assisted Network) is the load-bearing object: two independent 2-qubit circuits, each applying $R_Y(\alpha)$ rotations to both qubits and then a CNOT gate, read out through Pauli-Z expectation values to form a four-dimensional latent vector $F$, which a single linear layer with sigmoid maps to the conductivity image. Because the circuit has a small fixed size, the classical path is only one matrix multiplication, so parameter count and runtime stop scaling with the number of pixels or voxels; the circuit's parameters act as a trainable nonlinear prior that is optimized jointly with the linear layer for each measurement.

What would settle it

Run the same pipeline with the two quantum circuits replaced by a classical four-parameter nonlinear feature map of the same functional form (for example, $f_i=\cos(\alpha_i)$ for two angles per circuit) with the same loss, regularization, and initialization; if that classical map recovers the bilateral lungs as accurately as QuantEIT on the 3D simulation, the claim that the quantum parameterization supplies indispensable prior information is refuted.

Watch

Extended reading notes

Core claim

The central claim is that the quantum circuit's output is not a substitute for the network but the network's essential core: the unknown conductivity change is parameterized as the output of a QA-Net made of two parallel 2-qubit circuits followed by one linear layer, $\Delta\sigma = \mathrm{Sigmoid}(W F + b)$, with $F\in\mathbb{R}^4$ assembled from Pauli-Z expectation values. During reconstruction, only the circuit rotation angles and the linear weights are optimized, and the linearized EIT objective $\|\Delta v - J\Delta\sigma\|_2^2$ plus Laplacian, total-variation, and $\ell^1$ regularizers provides the supervision. The paper reports that this configuration matches or beats the traditional one-step reconstructor and the state-of-the-art unsupervised shallow-network baseline on simulated, phantom, and clinical 2D and 3D lung data, and that swapping the quantum outputs for a fixed or learnable four-vector makes reconstruction fail, which it reads as evidence that the circuit's entanglement and nonlinearity are an indispensable prior.

Load-bearing premise

The load-bearing premise is that the particular values produced by the quantum circuits, not just the fact that four numbers are fed into a linear layer, are essential, so that a classical four-parameter feature map producing the same values would not reconstruct equally well.

Editorial extensions

If this is right

  • QuantEIT, if the reported comparisons hold, reconstructs 2D and 3D EIT images with roughly 0.2% of the parameters of the R-SIP baseline while staying within about 10% of its best quality scores even at 10–20 dB noise.
  • The unsupervised, training-data-free formulation removes the need for large labeled EIT datasets and should transfer to unseen anatomies and imaging conditions without retraining.
  • Because the quantum circuit's size is fixed, inference time stays nearly constant as the output resolution grows, unlike fully connected classical reconstructions whose cost rises sharply in 3D.
  • The reported ablation implies that the specific nonlinear structure of the circuit output, and not merely the dimensionality of the latent vector, determines whether the ill-posed inversion succeeds.
  • The framework is presented as the first integration of quantum circuits into EIT image reconstruction and is intended as a step toward running the same reconstruction on quantum hardware.

Reading between the lines

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

  • The paper leaves open whether a classical four-parameter nonlinear feature map with the same form, such as $f_i=\cos(\alpha_i)$ or a tiny random MLP, would reconstruct just as well; if one does, the quantum parameterization is a convenience rather than a necessity.
  • Scaling the latent space by increasing the number of qubits or the number of parallel circuits could reveal how reconstruction quality depends on circuit expressivity, a test the paper does not run.
  • The same recipe—a fixed small quantum circuit generating a latent prior plus one linear layer inside an untrained data-fidelity loop—could be carried over to other ill-posed imaging problems such as CT or positron emission tomography, which the paper names as future work.
  • Running the circuits on an actual noisy quantum device would test whether the reported robustness to measurement noise survives gate noise and finite measurement shots, a regime that classical simulation cannot probe.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 7 minor

Summary. The manuscript proposes QuantEIT, an unsupervised hybrid quantum-classical framework for electrical impedance tomography (EIT). Two parallel two-qubit parameterized circuits (RY rotations and CNOT entanglement) produce a four-dimensional feature vector F; a single linear layer maps F to a conductivity image via Δσ = Sigmoid(WF + b). For each measurement, the parameters are optimized to fit the linearized forward model JΔσ to normalized voltage changes, with Laplacian, total-variation, and ℓ1 regularizers. The authors report experiments on 2D and 3D simulations, a lung phantom, and three human participants, comparing against Noser and R-SIP, and claim comparable or superior reconstruction accuracy with about 0.2% of R-SIP's parameters and enhanced noise robustness. An ablation replacing the circuit outputs with fixed or learnable 4-vectors is presented as evidence that the quantum circuit supplies essential prior information.

Significance. If the central claim were established, QuantEIT would be a notable contribution: it is a genuinely small per-image prior for EIT, it appears to be the first quantum-circuit-based EIT reconstruction, and the clinical and phantom validation goes beyond typical simulation-only studies. The paper is clearly written, uses a standard linearized forward model and standard metrics, and includes real-world data collected under IRB approval. However, the quantum benefit rests on an ablation that is logically incapable of supporting it, because Eq. (13) gives all compared variants the same expressive image set. The paper also lacks statistical replication, and its wall-clock inference times undermine the efficiency claim. The experimental infrastructure is good, but the novelty claim is currently unsupported; with additional controls and appropriate repositioning, the work could become publishable.

major comments (3)
  1. [IV-A3 (Ablation Study), Eq. (13)] The ablation cannot support the claim that entanglement and nonlinearity provide essential prior information. For any fixed nonzero F, the affine map (W,b) -> WF + b is surjective onto R^n: given any target pre-activation z, set W_i = z_i F / ||F||^2 and b = 0. Consequently, Δσ = Sigmoid(WF + b) has exactly the same image (0,1)^n for constant, learnable, and quantum-generated F; all three variants are equally expressive. The reported failure of the two non-quantum controls must therefore be an optimization artifact (initialization, learning-rate/regularizer interaction, or premature convergence), not a lack of representational capacity. To substantiate the quantum-prior claim, the authors should compare against a classical 4-parameter nonlinear feature map with the same image (e.g., (cos φ1, sin φ1, cos φ2, sin φ2)), report multiple random restarts with loss curves, and demonstrate that the circuit parameterization changes the optimization landscape rather than the final image set.
  2. [IV-A (Figs. 7-8)] The quantitative superiority claims are based on single runs without error bars. The 2D and 3D simulation results report one CC, PSNR, ERR, and MSSIM value per method, with no standard deviations, repeated random initializations, or multiple noise realizations for the main comparison. Since the method is stochastic (random initialization and Adam optimization) and per-image, the claim that QuantEIT outperforms R-SIP across metrics is not statistically substantiated. Please add mean and standard deviation over at least 5-10 restarts or noise realizations, and ideally multiple anatomical models, before claiming consistent superiority.
  3. [IV-A4 (Computational Complexity and Inference Time), Fig. 10, Table I] The efficiency claim is incomplete and partially contradicted by the reported inference times. QuantEIT takes 18.14 s for 2D and 19.05 s for 3D reconstruction, while R-SIP takes 5.02 s for 2D and 32.33 s for 3D. The paper highlights negligible FLOPs and 0.2% of the parameters, but for a bedside monitoring application with high temporal resolution, total end-to-end reconstruction time is the relevant quantity, and QuantEIT is not faster in the 2D case. Please report end-to-end per-image reconstruction time for all methods, discuss the speed-parameter trade-off explicitly, and avoid the phrase efficient inference if the contribution is parameter efficiency rather than wall-clock speed.
minor comments (7)
  1. [Fig. 1] In the circuit diagram, the second quantum circuit appears to label both rotations as α0^(2); presumably the second rotation should be α1^(2). Please correct the figure to match Eq. (5)-(7).
  2. [II-B, Eq. (11)] The index i is used both for qubits within a circuit and for the parallel circuits; please clarify the notation so that the qubit index and the circuit index are distinguished.
  3. [IV-A1] The statement that performance degradation remains within approximately 10% under strong noise is vague; specify the reference metric and report the actual CC, PSNR, ERR, or MSSIM values at 10 dB and 20 dB.
  4. [Fig. 8] The quantitative values for Noser and R-SIP are only visible in figures and are not given in the text or a table; reporting exact numbers would improve reproducibility and allow readers to assess the margins of improvement.
  5. [III-E] Several hyperparameters (λ vector, learning rate, Noser μ) are selected based on trial and error or convergence curves; a sensitivity analysis, or at least a statement of how robust the conclusions are to these choices, would strengthen the paper, especially because the ablation outcome may depend on them.
  6. [General] The manuscript does not include a data or code availability statement. For a study with clinical and quantitative claims, releasing anonymized data and code would substantially improve reproducibility.
  7. [Abstract and Conclusion] The claim of using only 0.2% of the parameters is a rounded approximation: the 2D ratio is about 0.23% and the 3D ratio is about 0.25%. The rounding is acceptable, but the text should say approximately 0.2%.

Circularity Check

1 steps flagged · score 6.0 of 10

The Sec. IV-A3 ablation cannot establish that the quantum latent is indispensable: by Eq. (13), W and b alone realize any output, so the constant and learnable control latents have identical representational power; the reported failure is an optimization artifact, not evidence of a quantum prior.

  1. other [Sec. II-C, Eq. (13); Sec. IV-A3]
    "∆σσσ = Sigmoid(WWW F + bbb), (13); These results demonstrate that the inherent entanglement and non-linearity of the quantum-assisted latent space provide essential prior information that cannot be substituted by either static or trainable embeddings."

    In Eq. (13), for any fixed nonzero F, the map (W,b) ↦ W F + b is surjective onto R^n: choosing W=0 and b=logit(t) realizes any target t ∈ (0,1)^n. Hence the constant-vector, learnable-vector, and quantum-circuit variants all have exactly the same set of representable conductivity images. The Sec. IV-A3 result that the non-quantum controls 'failed to reconstruct any meaningful 3D structures' therefore cannot be evidence that the quantum latent provides 'essential prior information'; it is an optimization artifact, not a property of the circuit. The claim that the quantum embedding 'cannot be substituted' is contradicted by the model's own construction, since the bias b alone can substitute for F.

full rationale

The main reconstruction comparisons against Noser and R-SIP are externally benchmarked and not circular: all methods are per-image fitting procedures, and reporting accuracy against ground truth for simulated data is standard evaluation of an inverse solver. The quantum-specific novelty, however, rests on the Sec. IV-A3 ablation, which is mathematically uncontrolled. Because Eq. (13) is an affine map of F followed by a sigmoid, and the bias vector b is free, the output distribution is the same whether F is a quantum-circuit output, a fixed vector, or a directly learned vector. The ablation's failure of the non-quantum controls therefore cannot be attributed to missing expressivity or missing prior information; it must stem from optimization dynamics (initialization, learning rate, or regularization interplay) that the paper does not analyze. This makes the central claim that entanglement/nonlinearity supplies an indispensable prior a claim that reduces, by the paper's own equations, to an uncontrolled fitting artifact. The other cited prior work (e.g., R-SIP) is used as a baseline, not as a load-bearing justification, so self-citation is not the issue. Overall, the framework may still be a useful lightweight per-image reconstruction method, but the quantum-necessity conclusion is partially circular by construction.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The method relies on the linearized EIT model, hand-tuned regularizers, and the unexplained choice of a 2-qubit circuit; the per-image optimization means all reported metrics are fits to the evaluated data, and no external predictive benchmark is provided. No new physical entities are introduced.

free parameters (7)
  • Quantum rotation angles phi (4 parameters) = Optimized per image; no fixed value
    These are the trainable gate parameters in the two 2-qubit circuits; they are tuned by Adam to fit each measurement.
  • Linear layer weights W and biases b = 20,484 (2D) or 204,804 (3D) parameters per image
    The single linear layer maps the 4D latent vector to the conductivity image; optimized per image.
  • Regularization weights lambda_Laplacian, lambda_TV, lambda_l1 = [0.03, 0.002, 0.01] (2D sim); [0.05, 0.03, 0.1] (2D clinical); [0.001, 0.001, 0.001] (3D)
    Hand-tuned per scenario using trial and error (Section III.E); these weights materially affect the loss and hence the reconstruction.
  • Learning rate = 0.05 (2D), 0.1 (3D)
    Chosen from convergence curves for each scenario (Section III.E and IV.A.1).
  • Number of optimization iterations N = 1000
    Fixed for all experiments; no convergence criterion is used.
  • 2D max normalization = Applied to 2D reconstructions only
    The paper normalizes 2D reconstructions by their maximum value before computing metrics, citing scaling ambiguity (Section IV.A.2). This changes the reported 2D metrics but is not applied to 3D.
  • Noser regularization mu = 20
    Selected by trial and error for the baseline comparison (Section III.D); not part of QuantEIT but affects the reported baseline performance.
assumptions (4)
  • domain assumption The linearized EIT forward model delta_v = J delta_sigma (Eq. 1) is an accurate approximation for reconstruction across all simulated, phantom, and clinical scenarios.
    The entire loss function (Eq. 15) and all results depend on the Jacobian J computed from a reference conductivity; if the linearization is poor, the reconstructed images are not meaningful. The paper does not validate the linearization against the full nonlinear forward model.
  • domain assumption The handcrafted regularizers Laplacian, TV, and l1, with weights fixed per scenario, provide appropriate priors for the reconstructed conductivity.
    The reconstruction quality is sensitive to these weights, which are chosen by trial and error (Section III.E). The method's success depends on these priors being well matched to EIT lung imaging.
  • standard math The parameterized 2-qubit circuit's outputs are differentiable and can be optimized by gradient descent using the parameter-shift rule.
    This is a standard property of expectation values of Pauli operators with respect to parameterized gates (Section II.B); it is needed for the Adam updates in Eq. (16).
  • ad hoc to paper A 2-qubit circuit with nq=2 and nc=2 is sufficiently expressive to serve as the latent generator for 2D and 3D EIT reconstruction.
    The paper states this configuration balances expressivity and efficiency (Section II.C) but does not derive it or compare it with other circuit widths. The entire ultra-lightweight claim depends on this choice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of QuantEIT: Ultra-Lightweight Quantum-Assisted Inference for Chest Electrical Impedance Tomography." pith.science (2026). https://pith.science/paper/JZ4M2OT2

@misc{pith2026250714031,
  author       = {Pith},
  title        = {Pith review of: QuantEIT: Ultra-Lightweight Quantum-Assisted Inference for Chest Electrical Impedance Tomography},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JZ4M2OT2}},
  note         = {Machine review of arXiv:2507.14031}
}
read the original abstract

Electrical Impedance Tomography (EIT) is a non-invasive, low-cost bedside imaging modality with high temporal resolution, making it suitable for bedside monitoring. However, its inherently ill-posed inverse problem poses significant challenges for accurate image reconstruction. Deep learning (DL)-based approaches have shown promise but often rely on complex network architectures with a large number of parameters, limiting efficiency and scalability. Here, we propose an Ultra-Lightweight Quantum-Assisted Inference (QuantEIT) framework for EIT image reconstruction. QuantEIT leverages a Quantum-Assisted Network (QA-Net), combining parallel 2-qubit quantum circuits to generate expressive latent representations that serve as implicit nonlinear priors, followed by a single linear layer for conductivity reconstruction. This design drastically reduces model complexity and parameter number. Uniquely, QuantEIT operates in an unsupervised, training-data-free manner and represents the first integration of quantum circuits into EIT image reconstruction. Extensive experiments on simulated and real-world 2D and 3D EIT lung imaging data demonstrate that QuantEIT outperforms conventional methods, achieving comparable or superior reconstruction accuracy using only 0.2% of the parameters, with enhanced robustness to noise.

Figures

Figures reproduced from arXiv: 2507.14031 by the authors.

Figure 1
Figure 1. Overview of the hybrid quantum-classical framework for EIT imaging. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the 2D and 3D simulation setups. The left shows the 2D [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the lung phantom experiment setup. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: CT scans of Participant 2. (a) Sagittal view, (b) Axial view, and (c) [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Global impedance dynamics and thoracic CT images from Participant [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: Convergence and robustness analysis of QuantEIT. (a) Convergence [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 9
Figure 9. Figure 9: Ablation study of the proposed QuantEIT on 3D simulation. [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 7
Figure 7. Figure 7: Qualitative comparison of 2D (a) and 3D (b) EIT simulation [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Quantitative comparison of 2D (a) and 3D (b) EIT simulation [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 10
Figure 10. Figure 10: Scalability of inference time with respect to the number of [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: Qualitative comparison of EIT reconstruction results across different [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]
Figure 12
Figure 12. Figure 12: Qualitative comparison of EIT reconstruction results across different algorithms on 3D clinical data from participant 3. [PITH_FULL_IMAGE:figures/full_fig_p009_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

40 extracted references · 16 canonical work pages

  1. [1]

    Three- dimensional electrical impedance tomography,

    P. Metherall, D. C. Barber, R. H. Smallwood, and B. H. Brown, “Three- dimensional electrical impedance tomography,” Nature, vol. 380, no. 6574, pp. 509–512, 1996

    1996

  2. [2]

    Multi-modal eit imaging using lensfree and flexible impedance sensor,

    H. Fang, R. B. Liu, J. Sun, P. O. Bagnaninchi, Z. Liu, and Y . Yang, “Multi-modal eit imaging using lensfree and flexible impedance sensor,” in 2024 IEEE SENSORS . IEEE, 2024, pp. 1–4

    2024

  3. [3]

    Regional lung perfusion as determined by electrical impedance tomography in comparison with electron beam ct imaging,

    I. Frerichs, J. Hinz, P. Herrmann, G. Weisser, G. Hahn, M. Quintel, and G. Hellige, “Regional lung perfusion as determined by electrical impedance tomography in comparison with electron beam ct imaging,” IEEE transactions on medical imaging , vol. 21, no. 6, pp. 646–652, 2002

    2002

  4. [4]

    Eichler, J

    L. Eichler, J. Mueller, J. Grensemann, I. Frerichs, C. Z ¨ollner, and S. Kluge, “Lung aeration and ventilation after percutaneous tracheotomy measured by electrical impedance tomography in non-hypoxemic criti- cally ill patients: a prospective observational study,” Annals of intensive care, vol. 8, pp. 1–8, 2018

    work page 2018

  5. [5]

    Supervised descent learning for thoracic electrical impedance tomography,

    K. Zhang, R. Guo, M. Li, F. Yang, S. Xu, and A. Abubakar, “Supervised descent learning for thoracic electrical impedance tomography,” IEEE Transactions on Biomedical Engineering, vol. 68, no. 4, pp. 1360–1369, 2020. 10

    work page 2020

  6. [6]

    Advances in electrical impedance tomography-based brain imaging,

    X.-Y . Ke, W. Hou, Q. Huang, X. Hou, X.-Y . Bao, W.-X. Kong, C.-X. Li, Y .-Q. Qiu, S.-Y . Hu, and L.-H. Dong, “Advances in electrical impedance tomography-based brain imaging,” Military Medical Research , vol. 9, no. 1, p. 10, 2022

    work page 2022

  7. [7]

    Capacitively coupled electrical impedance tomography for brain imaging,

    Y . Jiang and M. Soleimani, “Capacitively coupled electrical impedance tomography for brain imaging,” IEEE transactions on medical imaging , vol. 38, no. 9, pp. 2104–2113, 2019

    work page 2019

  8. [8]

    Imaging fast electrical activity in the brain with electrical impedance tomography,

    K. Y . Aristovich, B. C. Packham, H. Koo, G. S. Dos Santos, A. McEvoy, and D. S. Holder, “Imaging fast electrical activity in the brain with electrical impedance tomography,” NeuroImage, vol. 124, pp. 204–213, 2016

    work page 2016

Show all 40 references
  1. [9]

    A digital twin of electrical tomography for quantitative multiphase flow imaging,

    S. Wang, D. Hu, M. Zhang, J. Qiu Lin, W. Chen, F. Giorgio-Serchi, L. Peng, Y . Li, and Y . Yang, “A digital twin of electrical tomography for quantitative multiphase flow imaging,” Communications Engineering , vol. 1, no. 1, p. 41, 2022

    2022

  2. [10]

    A virtual platform of electrical tomography for multiphase flow imaging,

    S.-N. Wang, F. Giorgio-Serchi, and Y .-J. Yang, “A virtual platform of electrical tomography for multiphase flow imaging,” Physics of Fluids , vol. 34, no. 10, 2022

    2022

  3. [11]

    Tikhonov regularization and prior information in electrical impedance tomography,

    M. Vauhkonen, D. Vad ´asz, P. A. Karjalainen, E. Somersalo, and J. P. Kaipio, “Tikhonov regularization and prior information in electrical impedance tomography,”IEEE transactions on medical imaging, vol. 17, no. 2, pp. 285–293, 1998

    1998

  4. [12]

    Noser: An algorithm for solving the inverse conductivity problem,

    M. Cheney, D. Isaacson, J. C. Newell, S. Simske, and J. Goble, “Noser: An algorithm for solving the inverse conductivity problem,” International Journal of Imaging systems and technology , vol. 2, no. 2, pp. 66–75, 1990

    1990

  5. [13]

    Adaptive l {p} regu- larization for electrical impedance tomography,

    J. Li, S. Yue, M. Ding, Z. Cui, and H. Wang, “Adaptive l {p} regu- larization for electrical impedance tomography,” IEEE Sensors Journal, vol. 19, no. 24, pp. 12 297–12 305, 2019

    2019

  6. [14]

    In vivo impedance imaging with total variation regularization,

    A. Borsic, B. M. Graham, A. Adler, and W. R. Lionheart, “In vivo impedance imaging with total variation regularization,” IEEE transac- tions on medical imaging , vol. 29, no. 1, pp. 44–54, 2009

    2009

  7. [15]

    Modl: Model-based deep learning architecture for inverse problems,

    H. K. Aggarwal, M. P. Mani, and M. Jacob, “Modl: Model-based deep learning architecture for inverse problems,” IEEE transactions on medical imaging, vol. 38, no. 2, pp. 394–405, 2018

    2018

  8. [16]

    Fista-net: Learning a fast iterative shrinkage thresholding network for inverse problems in imaging,

    J. Xiang, Y . Dong, and Y . Yang, “Fista-net: Learning a fast iterative shrinkage thresholding network for inverse problems in imaging,” IEEE Transactions on Medical Imaging , vol. 40, no. 5, pp. 1329–1339, 2021

    2021

  9. [17]

    Point- cloud transformer for 3-d electrical impedance tomography,

    Z. Chen, H. Zhang, D. Hu, C. Tan, Z. Liu, and Y . Yang, “Point- cloud transformer for 3-d electrical impedance tomography,” IEEE Transactions on Instrumentation Measurement , vol. 73, p. 3413161, 2024

    2024

  10. [18]

    Review of machine learning for bioimpedance tomography in regenerative medicine,

    Z. Liu, Z. Chen, and Y . Yang, “Review of machine learning for bioimpedance tomography in regenerative medicine,” in Diverse Per- spectives and State-of-the-Art Approaches to the Utilization of Data- Driven Clinical Decision Support Systems . IGI Global Scientific Publishing, ...

    2023

  11. [19]

    Deep d-bar: Real-time electrical impedance tomography imaging with deep neural networks,

    S. J. Hamilton and A. Hauptmann, “Deep d-bar: Real-time electrical impedance tomography imaging with deep neural networks,” IEEE transactions on medical imaging , vol. 37, no. 10, pp. 2367–2377, 2018

    2018

  12. [20]

    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 Journal , vol. 19, no. 1, pp. 196–204, 2018

    2018

  13. [21]

    Electrical resistance tomography image reconstruction with densely connected convolutional neural network,

    F. Li, C. Tan, and F. Dong, “Electrical resistance tomography image reconstruction with densely connected convolutional neural network,” IEEE Transactions on Instrumentation and Measurement , vol. 70, pp. 1–11, 2020

    2020

  14. [22]

    A learning- based method for solving ill-posed nonlinear inverse problems: A simulation study of lung eit,

    J. K. Seo, K. C. Kim, A. Jargal, K. Lee, and B. Harrach, “A learning- based method for solving ill-posed nonlinear inverse problems: A simulation study of lung eit,”SIAM journal on Imaging Sciences, vol. 12, no. 3, pp. 1275–1295, 2019

    2019

  15. [23]

    Electrical resistance tomography with conditional generative adversarial networks,

    Y . Chen, K. Li, and Y . Han, “Electrical resistance tomography with conditional generative adversarial networks,” Measurement Science and Technology, vol. 31, no. 5, p. 055401, 2020

    2020

  16. [24]

    Multi- frequency electrical impedance tomography reconstruction with multi- branch attention image prior,

    H. Fang, Z. Liu, Y . Feng, Z. Qiu, P. Bagnaninchi, and Y . Yang, “Multi- frequency electrical impedance tomography reconstruction with multi- branch attention image prior,” arXiv preprint arXiv:2409.10794 , 2024

    2024 arXiv

  17. [25]

    Mmv-net: A multi- ple measurement vector network for multifrequency electrical impedance tomography,

    Z. Chen, J. Xiang, P.-O. Bagnaninchi, and Y . Yang, “Mmv-net: A multi- ple measurement vector network for multifrequency electrical impedance tomography,” IEEE Transactions on Neural Networks and Learning Systems, vol. 34, no. 11, pp. 8938–8949, 2022

    2022

  18. [26]

    Deep image prior,

    D. Ulyanov, A. Vedaldi, and V . Lempitsky, “Deep image prior,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 9446–9454

    2018

  19. [27]

    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 Transactions on Pattern Analysis and Machine Intelligence , vol. 45, no. 8, pp. 9627–9638, 2023

    2023

  20. [28]

    Regular- ized shallow image prior for electrical impedance tomography,

    Z. Liu, Z. Chen, H. Fang, Q. Wang, S. Zhang, and Y . Yang, “Regular- ized shallow image prior for electrical impedance tomography,” IEEE Transactions on Instrumentation and Measurement , 2025

    2025

  21. [29]

    The complexity of nisq,

    S. Chen, J. Cotler, H.-Y . Huang, and J. Li, “The complexity of nisq,” Nature Communications, vol. 14, no. 1, p. 6001, 2023

    2023

  22. [30]

    Deep quantum neural networks on a superconducting processor,

    X. Pan, Z. Lu, W. Wang, Z. Hua, Y . Xu, W. Li, W. Cai, X. Li, H. Wang, Y .-P. Songet al., “Deep quantum neural networks on a superconducting processor,” Nature Communications, vol. 14, no. 1, p. 4006, 2023

    2023

  23. [31]

    Quantum neural networks: A comparative analysis and noise robustness evaluation,

    T. Ahmed, M. Kashif, A. Marchisio, and M. Shafique, “Quantum neural networks: A comparative analysis and noise robustness evaluation,” arXiv preprint arXiv:2501.14412 , 2025

    2025 arXiv

  24. [32]

    Quantum convolutional neural network for classical data classification,

    T. Hur, L. Kim, and D. K. Park, “Quantum convolutional neural network for classical data classification,” Quantum Machine Intelligence , vol. 4, no. 1, p. 3, 2022

    2022

  25. [33]

    Generalization in quantum machine learning from few training data,

    M. C. Caro, H.-Y . Huang, M. Cerezo, K. Sharma, A. Sornborger, L. Cincio, and P. J. Coles, “Generalization in quantum machine learning from few training data,” Nature communications, vol. 13, no. 1, p. 4919, 2022

    2022

  26. [34]

    Pennylane: Automatic differentiation of hybrid quantum-classical com- putations,

    V . Bergholm, J. Izaac, M. Schuld, C. Gogolin, S. Ahmed, V . Ajith, M. S. Alam, G. Alonso-Linaje, B. AkashNarayanan, A. Asadi et al. , “Pennylane: Automatic differentiation of hybrid quantum-classical com- putations,” arXiv preprint arXiv:1811.04968 , 2018

    2018 arXiv

  27. [35]

    M. A. Nielsen and I. L. Chuang, Quantum computation and quantum information. Cambridge university press, 2010

    2010

  28. [36]

    Adam: A method for stochastic optimization,

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

    2014 arXiv

  29. [37]

    P. C. Hansen, Rank-deficient and discrete ill-posed problems: numerical aspects of linear inversion . SIAM, 1998

    1998

  30. [38]

    Nonlinear total variation based noise removal algorithms,

    L. I. Rudin, S. Osher, and E. Fatemi, “Nonlinear total variation based noise removal algorithms,” Physica D: nonlinear phenomena , vol. 60, no. 1-4, pp. 259–268, 1992

    1992

  31. [39]

    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,” inProceedings of the 6th Conference on Biomedical Applications of Electrical Impedance Tomography, London , 2005

    2005

  32. [40]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,” IEEE transactions on image processing , vol. 13, no. 4, pp. 600–612, 2004

    2004

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.