Pith. sign in

REVIEW 4 major objections 5 minor 79 references

Input layer regularization and automated regularization hyperparameter tuning for myelin water estimation using deep learning

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Pith's one-line read Input layer regularization—concatenating a noisy MRI decay signal with a Tikhonov-regularized version of itself—lowers myelin water fraction estimation error relative to plain networks and classical NLLS in simulation and in vivo brain…

desk verdict Solid synthetic evidence for ILR-based MWF estimation, but the in vivo comparison is undermined by an unaddressed echo-time grid mismatch. read the letter →

arxiv 2501.18074 v1 pith:SU34LBFV submitted 2025-01-30 q-bio.QM math.OCstat.APstat.COstat.ML

classification q-bio.QMmath.OCstat.APstat.COstat.ML MSC 93E2492B2045Q0562P1092C55
keywords InputlayerregularizationMyelinwaterfractionBiexponentialanalysisTikhonovGeneralizedcrossvalidationBileveloptimizationMagneticresonancerelaxometryDeeplearning
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 proposes a hybrid method, input layer regularization (ILR), that combines classical Tikhonov regularization with a neural network for estimating myelin water fraction from MRI biexponential decay signals. Instead of feeding the network only the measured noisy signal, ILR appends a regularized reconstruction of that same signal, with the per-signal regularization strength chosen either by generalized cross validation or by a trained network. On simulated data, the augmented networks estimate the fast-decay fraction $c_1$ with lower root mean squared error than a plain network, Tikhonov-regularized NLLS, and conventional NLLS at all tested SNR levels. On in vivo brain data, the ILR networks again outperform the plain network, and GCV-based selection is somewhat superior to network-based selection. This matters because myelin water fraction is a candidate biomarker for demyelinating disease, so more accurate pixel-wise estimation could improve clinical mapping of white matter.

What carries the argument

The load-bearing object is the augmented input vector $x=(s,\, G(p^*_{\lambda(s)}(s)))$, in which the noisy decay vector $s$ is concatenated with a regularized noiseless signal generated from a Tikhonov-regularized NLLS fit at a per-signal $\lambda(s)$. Two selectors produce $\lambda(s)$: a convolutional network $\lambda_{\mathrm{NN}}$ trained with $L^1$ loss to match an oracle $\lambda$ computed by grid search on the bilevel problem, and generalized cross validation $\lambda_{\mathrm{GCV}}$ for nonlinear inverse problems. The parameter-estimation network then maps this concatenated vector to $(c_1,T_{2,1},T_{2,2})$; a control network (ND, ND) receives the signal concatenated with itself to keep input length identical. The regularized component injects a smooth, denoised version of the signal so that the estimator can exploit both the raw measurements and a stabilized reconstruction.

What would settle it

Train both (ND, Reg) and (ND, ND) on the published grid, then evaluate them on known test signals with parameters outside that grid, for example $c_1>0.6$ or $T_{2,1}<1$ ms; if the ILR advantage disappears or reverses there, the improvement has only been shown inside the training distribution.

Watch

Extended reading notes

Core claim

The central claim is that appending a Tikhonov-regularized version of the noisy decay signal to the network input improves estimation of $c_1$, the myelin water fraction, in the three-parameter biexponential model $s(t)=c_1 e^{-t/T_{2,1}}+(1-c_1)e^{-t/T_{2,2}}$ under Rician noise. The regularized component is formed as $G(p^*_{\lambda(s)}(s))$, where $p^*_{\lambda(s)}$ is the Tikhonov-regularized NLLS estimate with a signal-dependent regularization parameter $\lambda(s)$ selected either by a convolutional network trained against an oracle $\lambda$ or by generalized cross validation. On synthetic testing data, both ILR variants reduce $c_1$ RMSE relative to the plain (ND, ND) network at SNR 5, 50, and 100, and both are far below NLLS and TR-NLLS; on human brain data, (ND, Reg) GCV outperforms (ND, Reg) NN and (ND, ND) across most pixels. The paper thereby claims that classical regularization-parameter selection can be productively embedded in a deep-learning parameter-estimation pipeline.

Load-bearing premise

The claimed accuracy gain assumes interpolation: every validation and test signal lies inside the parameter ranges used for training, and no out-of-distribution or extrapolation experiment is performed.

Editorial extensions

If this is right

  • The same ILR network outputs improved estimates of $T_{2,1}$ and $T_{2,2}$ along with $c_1$, so the method addresses the full biexponential parameter vector, not just the myelin water fraction.
  • Because $\lambda_{\mathrm{NN}}$ approximates the oracle distribution better by earth mover's distance while GCV better captures the low-SNR, small-$\lambda$ modes, the choice of regularizer selector can be guided by the signal's SNR regime.
  • Restricting analysis to AIC-selected biexponential pixels avoids the underdetermined monoexponential case, so ILR is intended for white-matter-like voxels rather than arbitrary tissue.
  • At medium and high SNR, ILR's advantage concentrates in the low-$c_1$ region with intermediate $T_{2,2}-T_{2,1}$ separation and in the roughly equal-weight region $c_1 \ge 0.4$ where the biexponential problem is most ill-posed.

Reading between the lines

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

  • Beyond the paper's experiments, ILR should transfer to other multiexponential signal models and to Gaussian noise, since the augmentation mechanism does not depend on the Rician likelihood or on the specific two-decay model.
  • This suggests an SNR-aware implementation: use $\lambda_{\mathrm{GCV}}$ when the estimated SNR is low and $\lambda_{\mathrm{NN}}$ when it is high, since their relative fidelity to the oracle distribution reverses with SNR.
  • A testable extension is to replace per-pixel independent processing with spatial context, for example by including neighboring voxels or a spatial penalty in the loss, which the paper notes as an open direction.
  • The parameter regions where the plain network wins—small $c_1$ with nearly identical decay times—imply ILR should be switched off when the fit approaches monoexponential behavior, rather than applied uniformly.
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

4 major / 5 minor

Summary. This manuscript proposes input layer regularization (ILR) for estimating the myelin water fraction c1 and the associated decay constants from biexponential MRI signals. The pipeline selects a Tikhonov regularization parameter per signal either by a dedicated neural network (λNN) or by generalized cross-validation (GCV), constructs a regularized signal G(p*_λ(s)), concatenates it with the noisy signal to form the network input, and trains a multilayer perceptron to estimate (c1, T2,1, T2,2). Synthetic experiments compare (ND,Reg)NN, (ND,Reg)GCV, (ND,ND), TR-NLLS, and NLLS at three SNR levels; in vivo brain data are analyzed after AIC-based selection of biexponential pixels, with a NESMA-denoised NLLS map as the reference standard. The central claim is that ILR significantly improves c1 estimation accuracy relative to plain networks and classical methods.

Significance. If the results hold, ILR is a simple and appealing hybrid method: it combines classical Tikhonov regularization with deep learning, removes the need for a manually fixed regularization parameter, and extends prior two-parameter work [36] to the practically relevant three-parameter problem including MWF. The manuscript's strengths include a reproducible synthetic design with known ground truth, comparison against several baselines, per-signal λ selection, and a detailed region-wise analysis in Appendix D. The main limitation is that the in vivo evaluation, which is essential for the practical claim, currently rests on a distribution-shifted input grid and a reference standard from the same model family as the method; the synthetic claims are also reported without uncertainty quantification.

major comments (4)
  1. [Appendix A, Table A.3; Appendix E.1] The synthetic training signals use Nt=32 acquisition times spanning [8.0, 256.0] ms, while the in vivo GRASE data acquire 32 echoes at t_n = n×11.3 ms, spanning [11.3, 361.6] ms. Because the network input is the raw signal vector with no explicit time-axis encoding, a change in the echo-time grid changes the input distribution. The text nowhere states that the in vivo signals were resampled to the training grid or that the networks were retrained on the in vivo grid. Consequently, the in vivo comparisons in Figure 7 and the conclusion that (ND,Reg) GCV outperforms (ND,ND) on brain data are out-of-distribution evaluations, and the claimed practical benefit is not yet supported. Please either demonstrate that the training grid covers the in vivo acquisition (for example, by resampling or retraining) or restrict the claims to the synthetic setting.
  2. [Table 2; Section 5] The central claim of a 'significant improvement' is based on single RMSE values per method and SNR, with no error bars, confidence intervals, or multiple-seed variability. At SNR=5 the difference between (ND,Reg)NN (0.1652) and (ND,ND) (0.1671) is about 1%, and at SNR=50 the difference between the two ILR variants is 0.0931 versus 0.0945. Without repeated training runs or a statistical test, the aggregate improvement at low SNR and the claimed difference between GCV and NN selection cannot be distinguished from noise. Please add uncertainty quantification or soften the significance claim accordingly.
  3. [Table A.3; Appendix D] The validation and testing parameter ranges (c1 in [0, 0.5], T2,1 in [5, 45], T2,2 in [45, 200]) are strictly inside the training ranges ([0, 0.6], [1, 50], [40, 225]), and no out-of-distribution or extrapolation experiments are reported. The reported accuracy gains are therefore interpolation results only. Since the in vivo data involve a different echo-time grid and likely different noise conditions, the paper should either add out-of-distribution tests (for example, parameters outside the training box, or synthetic signals on the in vivo grid) or explicitly limit the generalization claim.
  4. [Section E.2; Eq. (13)] The in vivo reference standard is obtained by applying the NESMA denoising filter followed by NLLS, and the ILR input in Eq. (13) contains a TR-NLLS-regularized curve. Both the reference and one input component therefore come from the same biexponential NLLS family, which could systematically favor (ND,Reg) over (ND,ND) in the in vivo comparison independent of any genuine ILR effect. The manuscript acknowledges that the reference is 'necessarily flawed' but does not address this shared-model concern. Please discuss this potential bias and, if possible, provide a validation on data whose ground truth is not NLLS-based (for example, synthetic data on the in vivo grid with known parameters).
minor comments (5)
  1. [Appendix D] In the low-SNR discussion, the sentence 'when 0.25 ≤ c1 ≤ 0.32, (ND, Reg) NN outperforms (ND, ND) for more values of (T21, T22) than (ND, Reg) NN' appears to compare (ND,Reg)NN with itself; it should likely compare (ND,Reg)NN with (ND,Reg)GCV.
  2. [Figure 7 caption] The right panel caption contains 'GCG', which should be 'GCV'.
  3. [Section 2.1] In the text after Eq. (1), 'the corresponding spin-spin decay constants T2,1 and T2,1' should read 'T2,1 and T2,2'.
  4. [Appendix B.2, Eq. (B.1)-(B.2)] Equations (B.1) and (B.2) contain unbalanced parentheses, and the loss multiplies the c1 error by 100 relative to the time constants; this weighting should be stated in the main text because it directly affects the reported RMSE comparisons.
  5. [Section 4.2 versus Appendix E.1] Section 4.2 states that 'a multi-spin-echo sequence with 64 values of TE was applied', whereas Appendix E.1 reports that echo data were acquired for 32 echoes with t_n = n×11.3 ms; these statements should be reconciled.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central ILR claim is evaluated against externally known synthetic ground truth, and the self-cited prior work is motivational rather than load-bearing.

full rationale

The central claim—that input layer regularization improves c1 estimation—is supported by Table 2, where all neural networks, including the ILR variants, are scored against held-out synthetic signals with known true parameters ptrue. This is an external benchmark, not a fitted quantity, so the synthetic comparison is not circular. The λoracle definition in Eq. 8 uses ptrue, but λoracle is explicitly an oracle comparison standard; the ILR networks at test time use λNN or λGCV, not λoracle, and their RMSE is computed against true parameters. The λNN is trained to reproduce λoracle, and Table 1 compares distributional fidelity of λNN versus λGCV to that oracle; this is a supervised fit-quality assessment, not a disguised prediction of the paper's own output. The self-citation of the prior ILR paper [36] is used as motivation ('ILR was shown to improve...'), but the present work re-implements and re-tests the method with new experiments, so the citation is not load-bearing for the current claim. The in vivo evaluation uses a NESMA-filtered NLLS surrogate for ground truth, and the ILR input includes a TR-NLLS-regularized curve; however, these are distinct estimators with different denoising and regularization procedures, so the comparison does not reduce by construction to the method's own inputs. The mismatch between synthetic and in vivo echo-time grids is a generalization or domain-shift concern, not circularity. Overall, no step in the derivation chain is equivalent to its own inputs by construction.

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

The method introduces no new physical entities. Its central claims rest on standard MRI and statistical assumptions, plus several hand-chosen parameters (network architecture, loss weights, parameter ranges, lambda grid, AIC selection). The narrow test ranges and the surrogate in vivo standard are the most consequential unverified choices.

free parameters (5)
  • Network architecture hyperparameters = Not fully specified
    Layer widths, kernel sizes, and depths for lambdaNN and PE-NNs are hand-chosen; the architecture is sketched in Figure 3 but numeric values are not given, and no systematic hyperparameter search is reported.
  • Loss weighting factor for c1 = 100.0
    Equation B.2 multiplies the c1 error by 100.0 while T2 errors are weighted by 1.0. This hand-chosen weighting shapes the training objective and directly affects reported RMSE values.
  • Regularization parameter search grid = lambda in [10^-7, 10^3]
    The grid range and density for computing lambda_oracle and lambda_GCV are chosen by hand and affect all lambda-dependent results.
  • Training, validation, and testing parameter ranges = c1: [0,0.6]; T2,1: [1,50]; T2,2: [40,225] training; narrower test ranges
    Table A.3 defines hand-chosen uniform grids and narrower test ranges, which make the evaluation an interpolation test and likely inflate accuracy.
  • AIC threshold for biexponential selection = None
    The choice to analyze only pixels with negative AIC difference (favoring biexponential) is a hand-set decision rule that determines the in vivo evaluation set.
assumptions (5)
  • domain assumption Magnitude MRI data follow a Rician noise model.
    Used in Eq. 4 and in all synthetic data generation (Algorithm 1). If the real noise deviates from Rician, the trained networks may not transfer.
  • domain assumption The bilexponential signal model with c2 = 1 - c1 is correct for myelin water imaging.
    The entire inverse problem, training data, and in vivo reference standard assume this three-parameter biexponential model (Eq. 2).
  • standard math The GCV formula for nonlinear inverse problems (Haber-Oldenburg) is valid for this problem.
    Equation 18 applies a nonlinear GCV criterion defined via the Jacobian J(lambda); the paper relies on this classical result without proof.
  • standard math Akaike information criterion with Gaussian errors is appropriate for selecting biexponential pixels.
    Appendix E.3 applies AIC to Rician-noise signals, assuming the noise is approximately Gaussian after NESMA filtering, which is an approximation.
  • domain assumption NESMA denoising followed by NLLS provides a trustworthy surrogate for true myelin water fraction in vivo.
    The in vivo reference standard is built from NESMA-filtered NLLS estimates (Section 4.2, Appendix E.2); this is acknowledged to be 'necessarily flawed' but is load-bearing for the in vivo comparison.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Input layer regularization and automated regularization hyperparameter tuning for myelin water estimation using deep learning." pith.science (2026). https://pith.science/paper/SU34LBFV

@misc{pith2026250118074,
  author       = {Pith},
  title        = {Pith review of: Input layer regularization and automated regularization hyperparameter tuning for myelin water estimation using deep learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SU34LBFV}},
  note         = {Machine review of arXiv:2501.18074}
}
read the original abstract

We propose a novel deep learning method which combines classical regularization with data augmentation for estimating myelin water fraction (MWF) in the brain via biexponential analysis. Our aim is to design an accurate deep learning technique for analysis of signals arising in magnetic resonance relaxometry. In particular, we study the biexponential model, one of the signal models used for MWF estimation. We greatly extend our previous work on \emph{input layer regularization (ILR)} in several ways. We now incorporate optimal regularization parameter selection via a dedicated neural network or generalized cross validation (GCV) on a signal-by-signal, or pixel-by-pixel, basis to form the augmented input signal, and now incorporate estimation of MWF, rather than just exponential time constants, into the analysis. On synthetically generated data, our proposed deep learning architecture outperformed both classical methods and a conventional multi-layer perceptron. On in vivo brain data, our architecture again outperformed other comparison methods, with GCV proving to be somewhat superior to a NN for regularization parameter selection. Thus, ILR improves estimation of MWF within the biexponential model. In addition, classical methods such as GCV may be combined with deep learning to optimize MWF imaging in the human brain.

Figures

Figures reproduced from arXiv: 2501.18074 by the authors.

Figure 1
Figure 1. The (ND, Reg) parameter estimation network is an NN where the inputs are the concatenated [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Combination of signal-dependent λ selection with input layer regularization. The first step involves selecting either a NN or GCV for estimating λ. Training data for the ILR networks (ND, Reg) are first passed through a TR-NLLS solver with the signal-dependent λ(s) to construct the concatenated vector x in Eq. (13). C(λ) is defined according to Eq. (18) 1D Convolution ReLU 1D Convolution ReLU Fully Connected ReLU Fu… view at source ↗
Figure 3
Figure 3. λNN architecture. 1-dimensional convolutional layers are applied to the input noisy signal, followed by fully connected layers to predict λoracle. C refers to the channel sizes (in/out for input or output channels, respectively). We also define the kernel size as Ksize and the padding as Psize. W refers to the width of a given fully-connected layer. The ReLU activation function is used between each layer. 8 [PITH_F… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: 3D Histogram showing prediction density of the [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Left to right: The true λ distribution (orange) overlaid with the distributions of λNN and λGCV . The true distributions at the low SNR level are more multimodal and have greater spread, with GCV providing superior estimates as compared to NN λ selection. This is parti…
Figure 6
Figure 6. Figure 6: Comparison standard for c1 for pixels determined to be biexponential according to the AIC (color bar); purple pixels are those determined to be of a monoexponential character. Each voxel in the 288 × 288 image of the brain is characterized by a noisy signal s with 64 a…
Figure 7
Figure 7. Figure 7: Left: Comparison of c1 estimation according to the two methods of λ selection. The metric illustrated is the difference in RMSE values RMSE((ND, Reg)NN) − RMSE((ND, Reg)GCV ) calculated over pixels with a biexponential character. As seen, (ND, Reg)GCV is superior overa…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

79 extracted references · 75 canonical work pages

  1. [36]

    Rozowski, J

    M. Rozowski, J. Palumbo, J. Bisen, C. Bi, M. Bouhrara, W. Czaja, R. G. Spencer, Input layer regularization for magnetic resonance relaxometry biexponential parameter estimation, Magnetic Resonance in Chemistry 60 (11) (2022) 1076–1086

  2. [67]

    Rozowski, J

    M. Rozowski, J. Palumbo, J. Bisen, C. Bi, M. Bouhrara, W. Czaja, R. G. Spencer, Input layer regularization for magnetic resonance relaxometry biexponential parameter estimation, Magn Reson Chem 60 (11) (2022) 1076–1086

  3. [1]

    T. D. Faizy, D. Kumar, G. Broocks, C. Thaler, F. Flottmann, H. Leischner, D. Kutzner, S. Hewera, D. Dotzauer, J.-P. Stellmann, et al., Age-related measurements of the myelin water fraction derived from 3d multi-echo grase reflect myelin content of the cerebral white matter, Scientific reports 8 (1) (2018) 14991

  4. [2]

    Laule, I

    C. Laule, I. M. Vavasour, G. R. Moore, J. Oger, D. K. Li, D. W. Paty, Water content and myelin water fraction in multiple sclerosis, Brain 127 (10) (2004) 2209–2218

  5. [3]

    A. N. Dula, D. F. Gochberg, H. L. Valentine, M. D. Does, Multiexponential t2, mag- netization transfer, and quantitative histology in white matter tracts of rat spinal cord, Magnetic Resonance in Medicine 63 (4) (2010) 902–909

  6. [4]

    MacKay, K

    A. MacKay, K. Whittall, J. Adler, D. Li, D. Paty, D. Graeb, In vivo visualization of myelin water in brain by magnetic resonance, Magnetic Resonance in Medicine 31 (6) (1994) 673–677

  7. [5]

    Kolind, L

    S. Kolind, L. Matthews, H. Johansen-Berg, M. I. Leite, S. C. Williams, S. Deoni, J. Palace, Myelin water imaging reflects clinical variability in multiple sclerosis, Neu- roimage 60 (1) (2012) 263–270

  8. [6]

    A. L. MacKay, C. Laule, Magnetic resonance of myelin water: an in vivo marker for myelin, Brain plasticity 2 (1) (2016) 71–91

Show all 79 references
  1. [7]

    I. M. Vavasour, C. Laule, D. K. Li, A. L. Traboulsee, A. L. MacKay, Is the magnetization transfer ratio a marker for myelin in multiple sclerosis?, Journal of Magnetic Resonance Imaging 24 (6) (2006) 809–814

  2. [8]

    S. C. Deoni, Quantitative relaxometry of the brain, Topics in Magnetic Resonance Imaging 21 (2) (2010) 101–113

  3. [9]

    J. Oh, E. T. Han, M. C. Lee, S. J. Nelson, D. Pelletier, Multi-component t2 relaxation and magnetization transfer in the normal brain and in cerebral gliomas, Journal of Neuroimaging 17 (4) (2007) 273–280

  4. [10]

    A. A. Istratov, O. F. Vyvenko, Exponential analysis in physical phenomena, Review of Scientific Instruments 70 (2) (1999) 1233–1257

  5. [11]

    Landaw, J

    E. Landaw, J. DiStefano 3rd, Multiexponential, multicompartmental, and noncompart- mental modeling. ii. data analysis and statistical considerations, American Journal of Physiology-Regulatory, Integrative and Comparative Physiology 246 (5) (1984) R665– R677

  6. [12]

    Ansorge, M

    R. Ansorge, M. Graves, The physics and mathematics of MRI, Morgan & Claypool Publishers, 2016

  7. [13]

    E.Alonso-Ortiz, I.R.Levesque, G.B.Pike, Mri-basedmyelinwaterimaging: atechnical review, Magnetic resonance in medicine 73 (1) (2015) 70–81. 27

  8. [14]

    Sijbers, A

    J. Sijbers, A. J. Den Dekker, P. Scheunders, D. Van Dyck, Maximum-likelihood esti- mation of rician distribution parameters, IEEE Transactions on Medical Imaging 17 (3) (1998) 357–361

  9. [15]

    Jiang, W

    L. Jiang, W. Yang, Adaptive magnetic resonance image denoising using mixture model and wavelet shrinkage, Dicta (2003) 831–838

  10. [16]

    R. C. Aster, B. Borchers, C. H. Thurber, Parameter estimation and inverse problems, Elsevier, 2018

  11. [17]

    Lanczos, Applied analysis, Courier Corporation, 1988

    C. Lanczos, Applied analysis, Courier Corporation, 1988

  12. [18]

    Calvetti, E

    D. Calvetti, E. Somersalo, Inverse problems: From regularization to bayesian inference, Wiley Interdisciplinary Reviews: Computational Statistics 10 (3) (2018) e1427

  13. [19]

    G. H. Golub, P. C. Hansen, D. P. O’Leary, Tikhonov regularization and total least squares, SIAM journal on matrix analysis and applications 21 (1) (1999) 185–194

  14. [20]

    Holler, K

    G. Holler, K. Kunisch, R. C. Barnard, A bilevel approach for parameter learning in inverse problems, Inverse Problems 34 (11) (2018) 115012

  15. [21]

    G. S. Alberti, E. De Vito, M. Lassas, L. Ratti, M. Santacesaria, Learning the optimal tikhonov regularizer for inverse problems, Advances in Neural Information Processing Systems 34 (2021) 25205–25216

  16. [22]

    Theodoridis, Machine learning: a Bayesian and optimization perspective, Academic press, 2015

    S. Theodoridis, Machine learning: a Bayesian and optimization perspective, Academic press, 2015

  17. [23]

    H. W. Engl, W. Grever, Using the l–curve for determining optimal regularization pa- rameters, Numerische Mathematik 69 (1) (1994) 25–31

  18. [24]

    M. T. Nair, U. Tautenhahn, E. Schock, Morozov’s discrepancy principle under general source conditions, Zeitschrift für Analysis und ihre Anwendungen 22 (1) (2003) 199–214

  19. [25]

    G. H. Golub, M. Heath, G. Wahba, Generalized cross-validation as a method for choos- ing a good ridge parameter, Technometrics 21 (2) (1979) 215–223

  20. [26]

    M. A. Lukas, Asymptotic optimality of generalized cross-validation for choosing the regularization parameter, Numerische Mathematik 66 (1993) 41–66

  21. [27]

    G.H.Golub, U.VonMatt, Generalizedcross-validationforlarge-scaleproblems, Journal of Computational and Graphical Statistics 6 (1) (1997) 1–34

  22. [28]

    Haber, D

    E. Haber, D. Oldenburg, A gcv based method for nonlinear ill-posed problems, Com- putational Geosciences 4 (2000) 41–63

  23. [29]

    B. M. Afkham, J. Chung, M. Chung, Learning regularization parameters of inverse problems via deep neural networks, Inverse Problems 37 (10) (2021) 105017. doi: 10.1088/1361-6420/ac245d. URL https://dx.doi.org/10.1088/1361-6420/ac245d 28

  24. [30]

    Kobler, A

    E. Kobler, A. Effland, K. Kunisch, T. Pock, Total deep variation for linear inverse problems, in: Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition, 2020, pp. 7549–7558

  25. [31]

    H. Li, J. Schwab, S. Antholzer, M. Haltmeier, Nett: Solving inverse problems with deep neural networks, Inverse Problems 36 (6) (2020) 065005

  26. [32]

    Ching, D

    T. Ching, D. S. Himmelstein, B. K. Beaulieu-Jones, A. A. Kalinin, B. T. Do, G. P. Way, E. Ferrero, P.-M. Agapow, M. Zietz, M. M. Hoffman, et al., Opportunities and obsta- cles for deep learning in biology and medicine, Journal of The Royal Society Interface 15 (141) (2018) 20170387

  27. [33]

    Saniei, S

    E. Saniei, S. Setayeshi, M. E. Akbari, M. Navid, Parameter estimation of breast tumour using dynamic neural network from thermal pattern, Journal of Advanced Research 7 (6) (2016) 1045–1055

  28. [34]

    S. G. Worswick, J. A. Spencer, G. Jeschke, I. Kuprov, Deep neural network processing of deer data, Science advances 4 (8) (2018) eaat5218

  29. [35]

    J. S. Almeida, E. O. Voit, Neural-network-based parameter estimation in s-system mod- els of biological networks, Genome Informatics 14 (2003) 114–123

  30. [37]

    Shorten, T

    C. Shorten, T. M. Khoshgoftaar, A survey on image data augmentation for deep learn- ing, Journal of big data 6 (1) (2019) 1–48

  31. [38]

    James, D

    G. James, D. Witten, T. Hastie, R. Tibshirani, et al., An introduction to statistical learning, Vol. 112, Springer, 2013

  32. [39]

    Bouhrara, D

    M. Bouhrara, D. A. Reiter, M. C. Maring, J.-M. Bonny, R. G. Spencer, Use of the nesma filter to improve myelin water fraction mapping with brain mri, Journal of neuroimaging 28 (6) (2018) 640–649

  33. [40]

    Zheng, H

    Y. Zheng, H. Chen, Q. Duan, L. Lin, Y. Shao, W. Wang, X. Wang, Y. Xu, Leveraging domain knowledge for robust deep reinforcement learning in networking, in: IEEE IN- FOCOM 2021-IEEE Conference on Computer Communications, IEEE, 2021, pp. 1–10

  34. [41]

    R. Guo, S. Xue, J. Hu, H. Sari, C. Mingels, K. Zeimpekis, G. Prenosil, Y. Wang, Y. Zhang, M. Viscione, et al., Using domain knowledge for robust and generalizable deep learning-based ct-free pet attenuation and scatter correction, Nature Communications 13 (1) (2022) 5882

  35. [42]

    Cuomo, V

    S. Cuomo, V. S. Di Cola, F. Giampaolo, G. Rozza, M. Raissi, F. Piccialli, Scientific machine learning through physics–informed neural networks: Where we are and what’s next, Journal of Scientific Computing 92 (3) (2022) 88. 29

  36. [43]

    S. Cai, Z. Mao, Z. Wang, M. Yin, G. E. Karniadakis, Physics-informed neural networks (pinns) for fluid mechanics: A review, Acta Mechanica Sinica 37 (12) (2021) 1727–1738

  37. [44]

    M. Zhou, J. Han, M. Rachh, C. Borges, A neural network warm-start approach for the inverse acoustic obstacle scattering problem, Journal of Computational Physics 490 (2023) 112341

  38. [45]

    Villar, D

    S. Villar, D. W. Hogg, K. Storey-Fisher, W. Yao, B. Blum-Smith, Scalars are univer- sal: Equivariant machine learning, structured like classical physics, Advances in Neural Information Processing Systems 34 (2021) 28848–28863

  39. [46]

    Burger, S

    M. Burger, S. Osher, Convergence rates of convex variational regularization, Inverse problems 20 (5) (2004) 1411

  40. [47]

    Lorenz, N

    D. Lorenz, N. Worliczek, Necessary conditions for variational regularization schemes, Inverse Problems 29 (7) (2013) 075016

  41. [48]

    Flemming, Theory and examples of variational regularization with non-metric fitting functionals (2010)

    J. Flemming, Theory and examples of variational regularization with non-metric fitting functionals (2010)

  42. [49]

    Scherzer, M

    O. Scherzer, M. Grasmair, H. Grossauer, M. Haltmeier, F. Lenzen, Variational methods in imaging, Vol. 167, Springer, 2009

  43. [50]

    Colson, P

    B. Colson, P. Marcotte, G. Savard, An overview of bilevel optimization, Annals of operations research 153 (2007) 235–256

  44. [51]

    Dempe, A

    S. Dempe, A. Zemkoho, Bilevel optimization, in: Springer optimization and its appli- cations, Vol. 161, Springer, 2020

  45. [52]

    Evgeniou, T

    T. Evgeniou, T. Poggio, M. Pontil, A. Verri, Regularization and statistical learning theory for data analysis, Computational Statistics & Data Analysis 38 (4) (2002) 421– 432

  46. [53]

    Cherkassky, Y

    V. Cherkassky, Y. Ma, Another look at statistical learning theory and regularization, Neural Networks 22 (7) (2009) 958–969

  47. [54]

    M. J. Ehrhardt, S. Gazzola, S. J. Scott, On optimal regularization parameters via bilevel learning, arXiv preprint arXiv:2305.18394 (2023)

  48. [55]

    A. Neubauer, An a posteriori parameter choice for tikhonov regularization in hilbert scales leading to optimal convergence rates, SIAM journal on numerical analysis 25 (6) (1988) 1313–1326

  49. [56]

    H. W. Engl, A. Neubauer, Optimal parameter choice for ordinary and iterated tikhonov regularization, in: Inverse and Ill-Posed Problems, Elsevier, 1987, pp. 97–125

  50. [57]

    O.Scherzer, H.W.Engl, K.Kunisch, Optimalaposterioriparameterchoicefortikhonov regularization for solving nonlinear ill-posed problems, SIAM journal on numerical anal- ysis 30 (6) (1993) 1796–1838. 30

  51. [58]

    S. H. Kolind, S. C. Deoni, Rapid simultaneous high-resolution mapping of myelin water fraction and relaxation times in human brain, NMR in Biomedicine 27 (5) (2014) 602– 614

  52. [59]

    MacKay, C

    A. MacKay, C. Laule, I. Vavasour, T. Bjarnason, S. Kolind, B. Mädler, D. K. Li, Insights into brain microstructure from the t2 distribution, Magnetic Resonance Imaging 24 (4) (2006) 515–525

  53. [60]

    Bouhrara, D

    M. Bouhrara, D. A. Reiter, C. M. Bergeron, L. M. Zukley, L. Ferrucci, S. M. Resnick, R. G. Spencer, Evidence of demyelination in mild cognitive impairment and demen- tia using a direct and specific magnetic resonance imaging measure of myelin content, Alzheimers Dement 14 (8) ...

  54. [61]

    D. A. Reiter, P.-C. Lin, K. W. Fishbein, R. G. Spencer, Multicomponent t2 relax- ation analysis in cartilage, Magnetic Resonance in Medicine: An Official Journal of the International Society for Magnetic Resonance in Medicine 61 (4) (2009) 803–809

  55. [62]

    Mossahebi, V

    P. Mossahebi, V. L. Yarnykh, A. Samsonov, Analysis and correction of biases in cross- relaxation mri due to biexponential longitudinal relaxation, Magnetic resonance in medicine 71 (2) (2014) 830–838

  56. [63]

    Gudbjartsson, S

    H. Gudbjartsson, S. Patz, The rician distribution of noisy mri data, Magnetic resonance in medicine 34 (6) (1995) 910–914

  57. [64]

    T. H. Edwards, S. Stoll, Optimal tikhonov regularization for deer spectroscopy, Journal of Magnetic Resonance 288 (2018) 58–68

  58. [65]

    S. Lunz, O. Öktem, C.-B. Schönlieb, Adversarial regularizers in inverse problems, Ad- vances in neural information processing systems 31 (2018)

  59. [66]

    C. Bi, M. Y. Ou, M. Bouhrara, R. G. Spencer, Span of regularization for solution of inverse problems with application to magnetic resonance relaxometry of the brain, Scientific reports 12 (1) (2022) 20194

  60. [68]

    Feurer, F

    M. Feurer, F. Hutter, Hyperparameter optimization, Automated machine learning: Methods, systems, challenges (2019) 3–33

  61. [69]

    Juras, S

    V. Juras, S. Apprich, Š. Zb` yň, L. Zak, X. Deligianni, P. Szomolanyi, O. Bieri, S. Trat- tnig, Quantitative mri analysis of menisci using biexponential t2* fitting with a variable echo time sequence, Magnetic resonance in medicine 71 (3) (2014) 1015–1023

  62. [70]

    D. A. Van Dyk, X.-L. Meng, The art of data augmentation, Journal of Computational and Graphical Statistics 10 (1) (2001) 1–50. 31

  63. [71]

    K. H. Chon, R. J. Cohen, Linear and nonlinear arma model parameter estimation using an artificial neural network, IEEE transactions on biomedical engineering 44 (3) (1997) 168–174

  64. [72]

    Calderón-Macías, M

    C. Calderón-Macías, M. K. Sen, P. L. Stoffa, Artificial neural networks for parameter estimation in geophysics [link], Geophysical prospecting 48 (1) (2000) 21–47

  65. [73]

    S. R. Green, C. Simpson, J. Gair, Gravitational-wave parameter estimation with au- toregressive neural network flows, Physical Review D 102 (10) (2020) 104057

  66. [74]

    J. N. Kutz, Machine learning for parameter estimation, Proceedings of the National Academy of Sciences 120 (12) (2023) e2300990120

  67. [75]

    M. D. DeVore, A. D. Lanterman, J. A. O’Sullivan, Atr performance of a rician model for sar images, in: Automatic target recognition X, Vol. 4050, SPIE, 2000, pp. 34–45

  68. [76]

    T. L. Marzetta, Em algorithm for estimating the parameters of a multivariate complex rician density for polarimetric sar, in: 1995 International Conference on Acoustics, Speech, and Signal Processing, Vol. 5, IEEE, 1995, pp. 3651–3654

  69. [77]

    C. G. Koay, P. J. Basser, Analytically exact correction scheme for signal extraction from noisy magnitude mr signals, Journal of magnetic resonance 179 (2) (2006) 317–322

  70. [78]

    Davoli, R

    E. Davoli, R. Ferreira, C. Kreisbeck, H. Schönberger, Structural changes in nonlocal denoising models arising through bi-level parameter learning, Applied Mathematics & Optimization 88 (1) (2023) 9

  71. [79]

    T. D. Faizy, C. Thaler, G. Broocks, F. Flottmann, H. Leischner, H. Kniep, J. Nawabi, G. Schön, J.-P. Stellmann, A. Kemmling, et al., The myelin water fraction serves as a marker for age-related myelin alterations in the cerebral white matter–a multiparametric mri aging study, ...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.