Pith. sign in

REVIEW 3 major objections 5 minor 34 references

Towards Globally Predictable k-Space Interpolation: A White-box Transformer Approach

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read A white-box Transformer for accelerated MRI is built by showing that the subgradient of a structured low-rank k-space model is exactly a multi-head subspace self-attention; the paper reports that this principled design outperforms CNN and b

desk verdict A genuinely new combination of SLR and attention for k-space, but the 'white-box' derivation is held together by an unproven approximation — send to review with the expectation of major revision. read the letter →

arxiv 2508.04051 v1 pith:WX7TGYXQ submitted 2025-08-06 cs.CV math.OC

classification cs.CVmath.OC
keywords k-spaceinterpolationwhite-boxTransformerstructuredlow-rankannihilatingfilterMRIreconstructionunfoldedoptimizationself-attentionsubgradient
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 claims that self-attention for k-space MRI interpolation is not a black-box choice but the natural subgradient of a structured low-rank model of global annihilation. If true, it means the attention mechanism can be derived from a principled objective rather than assumed, and the network becomes an unfolded optimization algorithm. The authors build GPI-WT by treating global annihilation filters as learnable matrices, approximating the subgradient of their SLR regularizer as softmax-based subspace attention, and unrolling gradient descent together with data consistency and a local predictability term. On knee MRI data, they report consistently better interpolation accuracy than CNN-based and black-box Transformer baselines under random and uniform masks at acceleration factors 4 and 6.

What carries the argument

The load-bearing object is the subgradient-to-softmax identity, Eq. (3): the matrix inverse $(I + \gamma (Q_h k)^*(Q_h k))^{-1}$ in the subgradient of the SLR penalty is replaced by $\operatorname{softmax}((Q_h k)^*Q_h k)$ (citing the sparse-rate-reduction derivation of white-box Transformers). This single step converts an optimization gradient into an attention head, with $Q_h$ simultaneously serving as query, key, and value. Around it, the paper relaxes the Hankel structure of the annihilation filters so $Q_h$ becomes freely learnable, then unrolls gradient descent into a cascade, using Swin-style window partitions to keep attention linear and a SPIRiT-derived linear kernel $G$ to retain l

What would settle it

Compute the exact subgradient of the Eq. (2) penalty for small random $k$ and $Q_h$ by eigenvalue decomposition and compare it with the right-hand side of Eq. (3). If the relative error is not small across a range of $Q_h$, the claimed identity between the SLR subgradient and the attention mechanism fails. A complementary check: in the trained GPI-WT network, replace softmax with the exact inverse $(I + \gamma (Q_h k)^*(Q_h k))^{-1}$ and see whether interpolation performance stays similar; a large change would show the network depends on a heuristic rather than the derived operation.

Watch

Extended reading notes

Core claim

The central claim is that the subgradient of the proposed structured low-rank regularizer, $R(k;Q^{[H]}) = \sum_{h=1}^{H}\operatorname{Tr}\ln\bigl(I + \gamma (Q_h k)^*(Q_h k)\bigr)$, is (up to the linear term $\gamma k$) exactly a multi-head subspace self-attention: each head computes $Q_h k\,\operatorname{softmax}((Q_h k)^*Q_h k)$. The same learnable matrix $Q_h$ plays the role of query, key, and value, and these matrices are the global annihilation filters of the low-rank model. Unrolling a gradient-descent iteration on the full objective---data consistency plus this regularizer plus a SPIRiT-based local predictability term---yields a cascaded network whose attention is a consequence of th

Load-bearing premise

The white-box claim rests on the unproven approximation that the matrix inverse $(I + \gamma (Q_h k)^*(Q_h k))^{-1}$ can be replaced by a softmax kernel (up to the $\gamma k$ term); if that approximation is inaccurate, the attention mechanism is not actually the subgradient of the stated low-rank model.

Editorial extensions

If this is right

  • If the subgradient identity holds, the attention maps in GPI-WT are interpretable as annihilation correlations, so reconstruction failures can be traced back to the low-rank model rather than to an opaque network.
  • Because the network is an unfolded optimization, its depth corresponds to iterations and its heads to blocks of the regularizer, giving principled ways to set depth, width, and regularization strength.
  • The ablation in the paper reports that both the linear-window attention and the local predictability term contribute, and that the white-box design beats a black-box Transformer with the same architecture.
  • The derivation provides a template for building interpretable Transformers for other inverse problems that admit annihilation or Hankel-structured low-rank models.

Reading between the lines

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

  • Extension: if the softmax replacement in Eq. (3) is treated as a design choice rather than a faithful derivation, one could substitute the exact inverse or other kernel normalizations and obtain a family of attention mechanisms with the same low-rank grounding; comparing them would test whether softmax is essential or merely convenient.
  • Extension: the linear-window attention points to non-local symmetries in k-space; a natural testable extension is whether trajectory-aware or coil-aware windowing further improves non-Cartesian and high-acceleration reconstructions.
  • Extension: the white-box claim is quantitatively checkable by measuring how close the trained network's attention is to the exact subgradient of the stated objective; reporting that residual would let readers verify the derivation rather than accept it on faith.
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 / 5 minor

Summary. The paper proposes GPI-WT, a k-space interpolation network for accelerated MRI that is claimed to be a white-box Transformer. The method introduces a structured low-rank (SLR) model with a log-determinant penalty on (Q_h k)^*(Q_h k), derives a subgradient, approximates it as multi-head subspace self-attention (MSSA), and unrolls a proximal gradient descent into a cascaded network. Experiments on knee MRI data compare the method against several k-space reconstruction baselines under random and uniform undersampling at AF=4 and AF=6, reporting improvements in NMSE, PSNR, and SSIM.

Significance. If the central derivation were correct, the paper would provide a principled connection between annihilation-based SLR priors and attention mechanisms, and the empirical gains would be of interest to the MRI reconstruction community. The paper also includes a systematic experimental comparison and ablations, which is a strength. However, the entire white-box claim rests on an unproven and mathematically questionable approximation in Eq. (3). Without a sound derivation, the contribution reduces to a heuristic attention architecture, and the claimed interpretability is not established.

major comments (3)
  1. [Section 2.1, Eq. (3)] The central step of the paper replaces (I + γ (Q_h k)^*(Q_h k))^{-1} by an expression involving softmax((Q_h k)^* Q_h k), citing [33]. This is not a justified approximation: no small-parameter condition, spectral norm bound, or error estimate is given. The cited reference addresses a different rate-reduction objective with different variables and normalizations, and does not directly apply to the log-determinant penalty in Eq. (2). Without this step, Eq. (4)–(5) do not follow from Eq. (2), and MSSA is not the subgradient of the proposed SLR model. This undermines the paper's central claim in the Abstract and Section 1.
  2. [Section 2.1, Eq. (3)] The approximation in Eq. (3) reads γ Σ_h Q_h^* Q_h k (I + γ A_h)^{-1} ≈ γk − γ^2 Σ_h Q_h^* Q_h k softmax(A_h). For the first term to become γk, one must have Σ_h Q_h^* Q_h = I (up to scaling). This condition is neither stated nor derivable from the SLR model, and it is not implied by the learnable Q_h. If it does not hold, the update rule in Eq. (7) is inconsistent with the gradient. Additionally, the gradient of log det(I + γ A) normally carries a factor 2 in real calculus; the convention is not clarified, and the omission affects the numerical constants in the network.
  3. [Section 2.1 and 2.2] The 'white-box' and 'interpretable' claims are over-stated. Even if Eq. (3) were valid, the learned parameters (Q_h, γ, λ1, λ2, μ, relative position bias B) are all data-driven, and the attention mechanism is constructed by choosing a regularizer whose subgradient, after approximation, resembles attention. The addition of the learnable bias B in Eq. (8) is a heuristic component not derived from the SLR model. Thus the network is at best 'inspired by' the SLR subgradient, not a faithful unfolding of a single principled objective. The paper should either justify this step rigorously or temper the white-box claim.
minor comments (5)
  1. [Throughout] There are numerous typos and inconsistencies: 'Globel' in Section 2.1 title, 'GPT-WT' vs 'GPI-WT' in Section 4 and Table 2/Figure 3, and missing spaces in the abstract. These should be corrected.
  2. [Section 2.1, Eq. (3)] The gradient formula should be stated with a clear convention for complex derivatives. The factor 2 discrepancy between the usual real gradient of log det(I + γ A) and the expression in Eq. (3) should be resolved, as it affects the interpretation of the step size μ.
  3. [Section 2.2] The use of square and linear window partitions limits attention to windows or lines, which is not truly 'global' in the sense of the title. The paper should clarify how the alternating window strategy approximates global dependencies and why it remains consistent with the 'globally predictable' model.
  4. [Section 3] Experimental details are incomplete: no information on batch size, number of epochs, data augmentation, coil sensitivity handling, or computational cost. The dataset is small (31 subjects training, 3 test) and no statistical significance tests are provided for the reported metric improvements.
  5. [References] The citation of [33] as the sole justification for the key approximation in Eq. (3) is insufficient and, as argued, likely incorrect. A direct derivation or a quantitative error bound is needed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the attention mechanism is derived from the chosen regularizer via an external approximation; the unproven approximation is a soundness risk, not a circular reduction.

full rationale

The derivation chain is linear: the paper defines the SLR regularizer R in Eq. (2), computes its subgradient exactly, then approximates the matrix inverse term with a softmax, explicitly citing [33]. The resulting expression is then named MSSA in Eqs. (4)-(5), and gradient descent on the full objective gives the unrolled network in Eq. (7). This is a mathematical derivation (modulo the cited approximation), not a case where a fitted parameter is renamed as a prediction or where the conclusion is assumed in the premise. The learnable filters Q_h are trained from data, but that is standard parameter fitting and does not make the derivation circular. The only load-bearing concern is the approximate equality in Eq. (3), which is not proved in the paper and may be invalid; however, that is a correctness or soundness gap, not a circular equivalence. The self-citations [2,18] are used for background context, not to justify the central derivation. Therefore no circular step is identified.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The central model relies on learned filters Q_h and a log-det penalty whose gradient is approximated as attention. The main free parameters are the Q_h matrices and several hyperparameters. The key axiom is the validity of the softmax approximation in Eq. (3), which is unproven. No new physical or abstract entities are introduced beyond the network components themselves.

free parameters (5)
  • Q_h (learned annihilation filters) = learned from training data
    The paper relaxes the Hankel structure and treats Q_h as learnable matrices, trained on a large dataset. The central SLR model depends on these filters, so the 'derivation' is not parameter-free.
  • γ (log-det penalty scale) = not specified
    Appears in the penalty and in the update step through the (1 - λ1 μ γ) factor. The paper does not state whether γ is fixed, learned, or tuned.
  • λ1, λ2, μ (regularization weights and step size) = not specified
    These control the balance between data consistency, global low-rank regularization, and local predictability. Their values are not reported, so they are hand-chosen or tuned.
  • Relative position bias B = learned
    Added to the attention in Eq. (8), increasing the number of learned parameters and influencing the attention pattern.
  • Architecture hyperparameters (T=10, window 4x4, H=6 heads) = T=10, 4x4, H=6
    Chosen empirically and not justified; they affect the network capacity and receptive field.
assumptions (3)
  • domain assumption There exist global annihilation filters Q_h such that Q_h k ≈ 0 for fully sampled k-space k.
    This is the structured low-rank assumption underlying Eq. (2). The paper states 'we consider the globally predictable and interpolated annihilation dependencies in k-space', but provides no evidence that such global filters exist for realistic MRI data.
  • ad hoc to paper The approximation (I + γ (Qk)^* (Qk))^{-1} ≈ softmax((Qk)^* (Qk)) is valid.
    Eq. (3) uses this approximate equality with a citation to [33], but the cited paper derives an approximation for a different objective (sparse rate reduction). No conditions or error bounds are given, yet the entire attention mechanism depends on it.
  • domain assumption The training dataset is representative of the test distribution.
    The model is trained on 31 subjects (840 slices) and tested on 3 subjects (96 slices) from the same scanner. Generalization to other scanners or contrasts is assumed but not tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Globally Predictable k-Space Interpolation: A White-box Transformer Approach." pith.science (2026). https://pith.science/paper/WX7TGYXQ

@misc{pith2026250804051,
  author       = {Pith},
  title        = {Pith review of: Towards Globally Predictable k-Space Interpolation: A White-box Transformer Approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WX7TGYXQ}},
  note         = {Machine review of arXiv:2508.04051}
}
read the original abstract

Interpolating missing data in k-space is essential for accelerating imaging. However, existing methods, including convolutional neural network-based deep learning, primarily exploit local predictability while overlooking the inherent global dependencies in k-space. Recently, Transformers have demonstrated remarkable success in natural language processing and image analysis due to their ability to capture long-range dependencies. This inspires the use of Transformers for k-space interpolation to better exploit its global structure. However, their lack of interpretability raises concerns regarding the reliability of interpolated data. To address this limitation, we propose GPI-WT, a white-box Transformer framework based on Globally Predictable Interpolation (GPI) for k-space. Specifically, we formulate GPI from the perspective of annihilation as a novel k-space structured low-rank (SLR) model. The global annihilation filters in the SLR model are treated as learnable parameters, and the subgradients of the SLR model naturally induce a learnable attention mechanism. By unfolding the subgradient-based optimization algorithm of SLR into a cascaded network, we construct the first white-box Transformer specifically designed for accelerated MRI. Experimental results demonstrate that the proposed method significantly outperforms state-of-the-art approaches in k-space interpolation accuracy while providing superior interpretability.

Figures

Figures reproduced from arXiv: 2508.04051 by the authors.

Figure 1
Figure 1. (a) Schematic of the our proposed unfolded network. (b) Diagram of window partitions. Smoothing filter w is applied to facilitate the computation in k-space. interpolation model, we derive a cascaded white-box Transformer network specif￾ically tailored for k-space interpolation. In particular, the (2)-regularized k-space interpolation model can be formulated as follows: min k 1 2 ∥MΩk − y∥ 2 2 + λ1R [PITH_FULL_IMAG… view at source ↗
Figure 2
Figure 2. Reconstruction results under random and uniform masks and AF = 4, 6 with 24 ACS lines. Values of NMSE(%)/PSNR/SSIM(%) are given. The second and third rows illustrate the enlarged and error views. Grayscale bars for reconstructed images and color bars for error maps are on figures’ right. significantly improves reconstruction quality by better capturing long-range de￾pendencies. Building upon this, the incorporation … view at source ↗
Figure 3
Figure 3. Ablation results under random mask at AF = 4 with 24 ACS lines. Values of NMSE(%)/PSNR/SSIM(%) are given. The second and third rows illustrate the enlarged and error views. Grayscale bars for reconstructed images and color bars for error maps are on figure’s right. Acknowledgments. This study was funded by Shenzhen Science and Technology Program under grant no. JCYJ20240813155840052; the National Key R&D Program of … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 29 canonical work pages

  1. [33]

    In: Proceedings of the 37th In- ternational Conference on Neural Information Processing Systems

    Yu, Y., Buchanan, S., Pai, D., Chu, T., Wu, Z., Tong, S., Haeffele, B.D., Ma, Y.: White-box transformers via sparse rate reduction. In: Proceedings of the 37th In- ternational Conference on Neural Information Processing Systems. NIPS ’23 (2023)

  2. [1]

    IEEE Transactions on Medical Imaging38(2), 394–405 (2019)

    Aggarwal, H.K., Mani, M.P., Jacob, M.: Modl: Model-based deep learning architec- ture for inverse problems. IEEE Transactions on Medical Imaging38(2), 394–405 (2019)

  3. [2]

    Medical Image Analysis88, 102877 (2023)

    Cui, Z.X., Jia, S., Cao, C., Zhu, Q., Liu, C., Qiu, Z., Liu, Y., Cheng, J., Wang, H., Zhu, Y., Liang, D.: K-unn: k-space interpolation with untrained neural network. Medical Image Analysis88, 102877 (2023)

  4. [3]

    In: International Conference on Learning Representations (2021)

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., Houlsby, N.: An image is worth 16x16 words: Transformers for image recognition at scale. In: International Conference on Learning Representations (2021)

  5. [4]

    Medical Image Analysis72, 102098 (2021) 10 C

    Du, T., Zhang, H., Li, Y., Pickup, S., Rosen, M., Zhou, R., Song, H.K., Fan, Y.: Adaptive convolutional neural networks for accelerating magnetic resonance imaging via k-space data interpolation. Medical Image Analysis72, 102098 (2021) 10 C. Luo et al

  6. [5]

    Magnetic Resonance in Medicine80(5), 2188–2201 (2018)

    Eo, T., Jun, Y., Kim, T., Jang, J., Lee, H.J., Hwang, D.: Kiki-net: cross-domain convolutionalneuralnetworksforreconstructingundersampledmagneticresonance images. Magnetic Resonance in Medicine80(5), 2188–2201 (2018)

  7. [6]

    Magnetic Resonance in Medicine47(6), 1202–1210 (2002)

    Griswold, M.A., Jakob, P.M., Heidemann, R.M., Nittka, M., Jellus, V., Wang, J., Kiefer, B., Haase, A.: Generalized autocalibrating partially parallel acquisitions (grappa). Magnetic Resonance in Medicine47(6), 1202–1210 (2002)

  8. [7]

    IEEE Transactions on Medical Imaging 43(1), 582–593 (2024)

    Guo, P., Mei, Y., Zhou, J., Jiang, S., Patel, V.M.: Reconformer: Accelerated mri reconstruction using recurrent transformer. IEEE Transactions on Medical Imaging 43(1), 582–593 (2024)

Show all 34 references
  1. [8]

    IEEE transactions on medical imaging33(3), 668–681 (2013)

    Haldar, J.P.: Low-rank modeling of localk-space neighborhoods (loraks) for con- strained mri. IEEE transactions on medical imaging33(3), 668–681 (2013)

  2. [9]

    Magnetic Resonance in Medicine 75(4), 1499–1514 (2016)

    Haldar, J.P., Zhuo, J.: P-loraks: Low-rank modeling of local k-space neighborhoods with parallel imaging data. Magnetic Resonance in Medicine 75(4), 1499–1514 (2016)

  3. [10]

    IEEE Transactions on Medical Imaging39(2), 377–386 (2020)

    Han, Y., Sunwoo, L., Ye, J.C.:k -space deep learning for accelerated mri. IEEE Transactions on Medical Imaging39(2), 377–386 (2020)

  4. [11]

    Mag- netic Resonance in Medicine6(1), 87–91 (1988)

    Hutchinson, M., Raff, U.: Fast mri data acquisition using multiple detectors. Mag- netic Resonance in Medicine6(1), 87–91 (1988)

  5. [12]

    IEEE Signal Processing Magazine 37(1), 54–68 (2020)

    Jacob, M., Mani, M.P., Ye, J.C.: Structured low-rank algorithms: Theory, magnetic resonance applications, and links to machine learning. IEEE Signal Processing Magazine 37(1), 54–68 (2020)

  6. [13]

    Kim, T.H., Garg, P., Haldar, J.P.: Loraki: Autocalibrated recurrent neu- ral networks for autoregressive mri reconstruction in k-space (2019), https://arxiv.org/abs/1904.09390

  7. [14]

    In: Proceedings of the International Conference on Learning Representations (2014)

    Kingma, D.P., Ba, J.: Adam: A method for stochastic optimization. In: Proceedings of the International Conference on Learning Representations (2014)

  8. [15]

    Magnetic Resonance in Medicine76(6), 1848–1864 (2016)

    Lee, D., Jin, K.H., Kim, E.Y., Park, S.H., Ye, J.C.: Acceleration of mr parameter mapping using annihilating filter-based low rank hankel matrix (aloha). Magnetic Resonance in Medicine76(6), 1848–1864 (2016)

  9. [16]

    Rev Magn Reson Med4(2), 67–185 (1992)

    Liang, Z.P., Boada, F., Constable, R., Haacke, E., Lauterbur, P., Smith, M.: Con- strained reconstruction methods in mr imaging. Rev Magn Reson Med4(2), 67–185 (1992)

  10. [17]

    In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV)

    Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., Guo, B.: Swin trans- former: Hierarchical vision transformer using shifted windows. In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV). pp. 9992–10002 (2021)

  11. [18]

    Luo, C., Wang, H., Xie, T., Jin, Q., Chen, G., Cui, Z.X., Liang, D.: Matrix completion-informed deep unfolded equilibrium models for self-supervisedk-space interpolation in mri (2023), https://doi.org/10.48550/arXiv.2309.13571

  12. [19]

    Magnetic Resonance in Medicine58(6), 1182–1195 (2007)

    Lustig, M., Donoho, D., Pauly, J.M.: Sparse mri: The application of compressed sensing for rapid mr imaging. Magnetic Resonance in Medicine58(6), 1182–1195 (2007)

  13. [20]

    IEEE Signal Processing Magazine25(2), 72–82 (2008)

    Lustig, M., Donoho, D.L., Santos, J.M., Pauly, J.M.: Compressed sensing mri. IEEE Signal Processing Magazine25(2), 72–82 (2008)

  14. [21]

    Magnetic Resonance in Medicine64(2), 457–471 (2010)

    Lustig, M., Pauly, J.M.: Spirit: Iterative self-consistent parallel imaging recon- struction from arbitrary k-space. Magnetic Resonance in Medicine64(2), 457–471 (2010)

  15. [22]

    IEEE Transactions on Medical Imaging39(12), 4186– 4197 (2020) Towards Globally Predictablek-Space Interpolation 11

    Pramanik, A., Aggarwal, H.K., Jacob, M.: Deep generalization of structured low- rank algorithms (deep-slr). IEEE Transactions on Medical Imaging39(12), 4186– 4197 (2020) Towards Globally Predictablek-Space Interpolation 11

  16. [23]

    Magnetic Resonance in Medicine 46(4), 638–651 (2001)

    Pruessmann, K.P., Weiger, M., Börnert, P., Boesiger, P.: Advances in sensitiv- ity encoding with arbitrary k-space trajectories. Magnetic Resonance in Medicine 46(4), 638–651 (2001)

  17. [24]

    Magnetic Resonance in Medicine42(5), 952–962 (1999)

    Pruessmann, K.P., Weiger, M., Scheidegger, M.B., Boesiger, P.: Sense: Sensitivity encoding for fast mri. Magnetic Resonance in Medicine42(5), 952–962 (1999)

  18. [25]

    In: Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomed- ical image segmentation. In: Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015. pp. 234–241 (2015)

  19. [26]

    Magnetic Resonance in Medicine72(4), 959–970 (2014)

    Shin, P.J., Larson, P.E.Z., Ohliger, M.A., Elad, M., Pauly, J.M., Vigneron, D.B., Lustig,M.:Calibrationlessparallelimagingreconstructionbasedonstructuredlow- rank matrix completion. Magnetic Resonance in Medicine72(4), 959–970 (2014)

  20. [27]

    IEEE Transactions on Medical Imaging5(3), 132–139 (1986)

    Smith, M.R., Nichols, S.T., Henkelman, R.M., Wood, M.L.: Application of au- toregressive moving average parametric modeling in magnetic resonance image reconstruction. IEEE Transactions on Medical Imaging5(3), 132–139 (1986)

  21. [28]

    Magnetic Resonance in Medicine 71(3), 990–1001 (2014)

    Uecker, M., Lai, P., Murphy, M.J., Virtue, P., Elad, M., Pauly, J.M., Vasanawala, S.S., Lustig, M.: Espirit—an eigenvalue approach to autocalibrating parallel mri: Where sense meets grappa. Magnetic Resonance in Medicine 71(3), 990–1001 (2014)

  22. [29]

    In: Proceedings of the 31st International Conference on Neural Information Processing Systems

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I.: Attention is all you need. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. p. 6000–6010. NIPS’17 (2017)

  23. [30]

    IEEE Transactions on Image Processing 13(4), 600–612 (2004)

    Wang, Z., Bovik, A., Sheikh, H., Simoncelli, E.: Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing 13(4), 600–612 (2004)

  24. [31]

    Computer Methods and Programs in Biomedicine 233, 107452 (2023)

    Wu, Z., Liao, W., Yan, C., Zhao, M., Liu, G., Ma, N., Li, X.: Deep learning based mri reconstruction with transformer. Computer Methods and Programs in Biomedicine 233, 107452 (2023)

  25. [32]

    SIAM Journal on Imaging Sciences11(2), 991– 1048 (2018)

    Ye, J.C., Han, Y., Cha, E.: Deep convolutional framelets: A general deep learning framework for inverse problems. SIAM Journal on Imaging Sciences11(2), 991– 1048 (2018)

  26. [34]

    In: British Machine Vision Conference (BMVC) (2022)

    Zhao, Z., Zhang, T., Xie, W., Wang, Y., Zhang, Y.: K-space transformer for un- dersampled mri reconstruction. In: British Machine Vision Conference (BMVC) (2022)

Pith tools

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