pith. machine review for the scientific record. sign in

arxiv: 2604.03150 · v1 · submitted 2026-04-03 · 💻 cs.LG

Recognition: 2 theorem links

· Lean Theorem

HyperFitS -- Hypernetwork Fitting Spectra for metabolic quantification of {}¹H MR spectroscopic imaging

Paul J. Weiser , Gulnur Ungan , Amirmohammad Shamaei , Georg Langs , Wolfgang Bogner , Malte Hoffmann , Antoine Klauser , Ovidiu C. Andronesi
Authors on Pith no claims yet

Pith reviewed 2026-05-13 19:35 UTC · model grok-4.3

classification 💻 cs.LG
keywords hypernetworkspectral fittingMRSImetabolite quantificationLCModeldeep learningmagnetic resonance spectroscopybaseline correction
0
0 comments X

The pith

HyperFitS uses a hypernetwork to adapt spectral fitting to different baselines and water suppression without retraining, matching LCModel results in seconds instead of hours.

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

The paper introduces HyperFitS, a hypernetwork for fitting proton MR spectra to quantify metabolites in whole-brain imaging. It dynamically adjusts to varying baseline corrections and water suppression factors by generating appropriate network weights on demand. This allows the same trained model to process data from multiple protocols and field strengths such as 3T and 7T without retraining. Metabolite maps agree closely with the conventional LCModel method while completing the task in seconds rather than hours. The work also demonstrates that baseline modeling choices can shift reported metabolite concentrations by as much as 30 percent.

Core claim

HyperFitS is a hypernetwork that produces weights for a main spectral-fitting network conditioned on baseline and water-suppression parameters. Applied to human 1H MRSI data at 3T and 7T with resolutions from 10 mm down to 2 mm, it yields metabolite concentration maps in substantial quantitative agreement with LCModel while reducing whole-brain processing time from hours to a few seconds. Unlike earlier neural fitting methods, the hypernetwork supports configurability across acquisition settings without new training runs.

What carries the argument

Hypernetwork that outputs weights for a spectral fitting network conditioned on baseline correction and water suppression parameters.

If this is right

  • Metabolite maps from HyperFitS show substantial agreement with LCModel across 3T and 7T acquisitions and multiple resolutions.
  • Whole-brain spectral fitting completes in seconds rather than the hours required by conventional methods.
  • Baseline parametrization choices can alter quantified metabolite levels by up to 30 percent.
  • The model handles both water-suppressed and water-unsuppressed data without retraining.
  • Configurability across protocols and field strengths is achieved through the hypernetwork without new training.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Near-real-time metabolite mapping could become practical during clinical scan sessions.
  • The same model might handle spectra from pathological tissue without retraining if baseline variability is the main change.
  • Integration into reconstruction pipelines could enable on-the-fly quantification for adaptive scanning.
  • Extension to other spectroscopic modalities with similar parameter variability becomes plausible.

Load-bearing premise

The hypernetwork can flexibly adapt its fitting to a broad range of baseline corrections and water suppression factors while maintaining accuracy comparable to LCModel across different field strengths and protocols without retraining.

What would settle it

MRSI spectra acquired with baseline distortions or water suppression factors far outside the tested range that produce metabolite values differing by more than 15 percent from LCModel fits on the same data would challenge the adaptability claim.

Figures

Figures reproduced from arXiv: 2604.03150 by Amirmohammad Shamaei, Antoine Klauser, Georg Langs, Gulnur Ungan, Malte Hoffmann, Ovidiu C. Andronesi, Paul J. Weiser, Wolfgang Bogner.

Figure 1
Figure 1. Figure 1: HyperFitS: Top: The HyperFitS strategy showing a hypernetwork tak [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: A qualitative comparison showing metabolic maps of NAA+NAAG, [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Metabolic maps of NAA+NAAG, Cr+PCr, GPC+PCh, Inositol, and [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Top: A representative spectrum fitted by varying the hyperparam￾eters for baseline flexibility (vertical) and spline node distance (horizontal) in increments of 1 10 of the total range. Three cases are shown enlarged on the right. The original spectrum is white, fit is red, baseline is yellow, and spline nodes are blue. Bottom: Contour plots of metabolite concentrations. A subject acquired at 7T with 3.4mm… view at source ↗
Figure 5
Figure 5. Figure 5: Metabolite concentration boxplots for metabolites NAA+NAAG, [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Uncertainty maps for the metabolites NAA, Glu and Ins at 7T (left) [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Correlation matrix of in-vivo metabolite concentrations fitted by Hy [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
read the original abstract

Purpose: Proton magnetic resonance spectroscopic imaging ($^1$H MRSI) enables the mapping of whole-brain metabolites concentrations in-vivo. However, a long-standing problem for its clinical applicability is the metabolic quantification, which can require extensive time for spectral fitting. Recently, deep learning methods have been able to provide whole-brain metabolic quantification in only a few seconds. However, neural network implementations often lack configurability and require retraining to change predefined parameter settings. Methods: We introduce HyperFitS, a hypernetwork for spectral fitting for metabolite quantification in whole-brain $^1$H MRSI that flexibly adapts to a broad range of baseline corrections and water suppression factors. Metabolite maps of human subjects acquired at 3T and 7T with isotropic resolutions of 10 mm, 3.4 mm and 2 mm by water-suppressed and water-unsuppressed MRSI were quantified with HyperFitS and compared to conventional LCModel fitting. Results: Metabolic maps show a substantial agreement between the new and gold-standard methods, with significantly faster fitting times by HyperFitS. Quantitative results further highlight the impact of baseline parametrization on metabolic quantification, which can alter results by up to 30%. Conclusion: HyperFitS shows strong agreement with state-of-the-art conventional methods, while reducing processing times from hours to a few seconds. Compared to prior deep learning based spectral fitting methods, HyperFitS enables a wide range of configurability and can adapt to data quality acquired with multiple protocols and field strengths without retraining.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces HyperFitS, a hypernetwork-based spectral fitting method for metabolite quantification in whole-brain 1H MRSI. It claims the approach flexibly adapts to a range of baseline corrections and water suppression factors without retraining, achieves substantial agreement with LCModel on human data acquired at 3T and 7T (isotropic resolutions 10 mm, 3.4 mm, 2 mm; water-suppressed and unsuppressed), and reduces fitting times from hours to seconds while highlighting up to 30% variation due to baseline parametrization.

Significance. If the central claims of cross-protocol, cross-field-strength adaptability hold with quantitative validation, the work would be significant for enabling rapid, configurable MRSI quantification in clinical settings, addressing a key practical barrier compared to conventional fitting and prior DL methods.

major comments (2)
  1. [Abstract/Methods] Abstract/Methods: The claim that a single trained hypernetwork adapts to 3T/7T data, multiple resolutions, and protocols without retraining is load-bearing but unsupported by details on whether field strength or resolution is provided as a conditioning input, the training distribution (mixed vs. single-field), or any held-out protocol test. The reported 30% baseline sensitivity makes unmodeled field-dependent effects (chemical-shift scaling, J-coupling, linewidth) a concrete risk to the zero-shot assertion.
  2. [Results] Results: The abstract states 'substantial agreement' with LCModel but supplies no quantitative metrics (correlation coefficients, mean absolute errors, sample sizes, error bars, or statistical tests), preventing assessment of whether the performance supports the cross-field and cross-protocol claims.
minor comments (2)
  1. Clarify the precise conditioning mechanism in the hypernetwork (e.g., how baseline correction parameters and water-suppression factors are encoded and injected).
  2. Add explicit dataset statistics (number of subjects, exact acquisition parameters per field strength/resolution) and any ablation studies on the hypernetwork's configurability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive review. The comments identify areas where additional methodological clarity and quantitative reporting will strengthen the manuscript. We address each major comment below and will revise the paper accordingly.

read point-by-point responses

  1. Referee: [Abstract/Methods] Abstract/Methods: The claim that a single trained hypernetwork adapts to 3T/7T data, multiple resolutions, and protocols without retraining is load-bearing but unsupported by details on whether field strength or resolution is provided as a conditioning input, the training distribution (mixed vs. single-field), or any held-out protocol test. The reported 30% baseline sensitivity makes unmodeled field-dependent effects (chemical-shift scaling, J-coupling, linewidth) a concrete risk to the zero-shot assertion.

    Authors: We agree that the manuscript requires more explicit description of the conditioning mechanism and training protocol to fully support the cross-field and cross-protocol claims. In the revised Methods section we will specify that field strength, voxel resolution, baseline correction parameters, and water-suppression factors are supplied as conditioning inputs to the hypernetwork. We will also document that the training set comprised a mixed distribution of 3T and 7T spectra acquired at the reported resolutions, and we will add results from held-out protocol evaluations. Regarding field-dependent effects, we will expand the discussion to explain how the hypernetwork explicitly conditions on linewidth and other protocol-specific parameters, thereby mitigating the risk highlighted by the 30% baseline sensitivity; we will further quantify this mitigation in the revised results. revision: yes

  2. Referee: [Results] Results: The abstract states 'substantial agreement' with LCModel but supplies no quantitative metrics (correlation coefficients, mean absolute errors, sample sizes, error bars, or statistical tests), preventing assessment of whether the performance supports the cross-field and cross-protocol claims.

    Authors: We concur that quantitative metrics are necessary to substantiate the agreement claims. Although the original manuscript contains visual metabolite maps and qualitative statements of agreement, we will revise the Results section and abstract to report Pearson correlation coefficients, mean absolute errors (or percentage errors), the number of subjects and voxels analyzed per field strength and resolution, error bars or confidence intervals, and appropriate statistical comparisons (e.g., Bland-Altman analysis). These additions will directly address the cross-field and cross-protocol performance. revision: yes

Circularity Check

0 steps flagged

No circularity: HyperFitS is an independently trained hypernetwork benchmarked against external LCModel results

full rationale

The paper presents a hypernetwork architecture whose weights are learned from training spectra (simulated or acquired) and whose outputs are compared to LCModel on separate test acquisitions at 3T/7T. No equation or claim reduces by construction to a fitted parameter being relabeled as a prediction, nor does any load-bearing step rely on a self-citation whose content is itself unverified. The configurability claim is an empirical statement about generalization across protocols, not a definitional identity. The derivation chain is therefore self-contained and externally falsifiable.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no information on training hyperparameters, network architecture details, loss functions, or data assumptions; ledger entries cannot be populated.

pith-pipeline@v0.9.0 · 5610 in / 1132 out tokens · 41724 ms · 2026-05-13T19:35:22.563326+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages · 1 internal anchor

  1. [1]

    Eccentric: a fast and unrestrained approach for high-resolution in vivo metabolic imaging at ultra-high field mr.Imaging Neuroscience, 2:1–20, 2024

    Antoine Klauser, Bernhard Strasser, Wolfgang Bogner, Lukas Hingerl, Se- bastien Courvoisier, Claudiu Schirda, Bruce R Rosen, Francois Lazeyras, and Ovidiu C Andronesi. Eccentric: a fast and unrestrained approach for high-resolution in vivo metabolic imaging at ultra-high field mr.Imaging Neuroscience, 2:1–20, 2024

    work page 2024

  2. [2]

    High-resolution metabolic imaging of high-grade gliomas using 7t-crt-fid-mrsi.NeuroImage: Clinical, 28:102433, 2020

    Gilbert Hangel, Cornelius Cadrien, Philipp Lazen, Julia Furtner, Alexandra Lipka, Eva Heˇ ckov´ a, Lukas Hingerl, Stanislav Motyka, Stephan Gruber, Bernhard Strasser, et al. High-resolution metabolic imaging of high-grade gliomas using 7t-crt-fid-mrsi.NeuroImage: Clinical, 28:102433, 2020. 17

    work page 2020

  3. [3]

    High-resolution 1h-mrsi of the brain using spice: data acquisition and image reconstruction.Magnetic resonance in medicine, 76(4):1059–1070, 2016

    Fan Lam, Chao Ma, Bryan Clifford, Curtis L Johnson, and Zhi-Pei Liang. High-resolution 1h-mrsi of the brain using spice: data acquisition and image reconstruction.Magnetic resonance in medicine, 76(4):1059–1070, 2016

    work page 2016

  4. [4]

    Methodological consensus on clinical proton mrs of the brain: Review and recommendations.Magnetic resonance in medicine, 82(2):527–550, 2019

    Martin Wilson, Ovidiu Andronesi, Peter B Barker, Robert Bartha, Alberto Bizzi, Patrick J Bolan, Kevin M Brindle, In-Young Choi, Cristina Cudalbu, Ulrike Dydak, et al. Methodological consensus on clinical proton mrs of the brain: Review and recommendations.Magnetic resonance in medicine, 82(2):527–550, 2019

    work page 2019

  5. [5]

    Estimation of metabolite concentrations from localized in vivo proton nmr spectra.Magnetic resonance in medicine, 30(6):672–679, 1993

    Stephen W Provencher. Estimation of metabolite concentrations from localized in vivo proton nmr spectra.Magnetic resonance in medicine, 30(6):672–679, 1993

    work page 1993

  6. [6]

    Preprocessing, analysis and quantification in single-voxel mag- netic resonance spectroscopy: experts’ consensus recommendations.NMR in Biomedicine, 34(5):e4257, 2021

    Jamie Near, Ashley D Harris, Christoph Juchem, Roland Kreis, Ma lgorzata Marja´ nska, G¨ ulin¨Oz, Johannes Slotboom, Martin Wilson, and Charles Gasparovic. Preprocessing, analysis and quantification in single-voxel mag- netic resonance spectroscopy: experts’ consensus recommendations.NMR in Biomedicine, 34(5):e4257, 2021

    work page 2021

  7. [7]

    H Ratiney, Micha¨ el Sdika, Y Coenradie, S Cavassila, D van Ormondt, and D Graveron-Demilly. Time-domain semi-parametric estimation based on a metabolite basis set.NMR in Biomedicine: An International Journal Devoted to the Development and Application of Magnetic Resonance In vivo, 18(1):1–13, 2005

    work page 2005

  8. [8]

    A constrained least-squares approach to the au- tomated quantitation of in vivo 1h magnetic resonance spectroscopy data

    Martin Wilson, Greg Reynolds, Risto A Kauppinen, Theodoros N Arvani- tis, and Andrew C Peet. A constrained least-squares approach to the au- tomated quantitation of in vivo 1h magnetic resonance spectroscopy data. Magnetic resonance in medicine, 65(1):1–12, 2011

    work page 2011

  9. [9]

    Fsl-mrs: An end- to-end spectroscopy analysis package.Magnetic resonance in medicine, 85(6):2950–2964, 2021

    William T Clarke, Charlotte J Stagg, and Saad Jbabdi. Fsl-mrs: An end- to-end spectroscopy analysis package.Magnetic resonance in medicine, 85(6):2950–2964, 2021

    work page 2021

  10. [10]

    Vespa: integrated applications for rf pulse design, spectral simulation and mrs data analysis.Magnetic resonance in medicine, 90(3):823–838, 2023

    Brian J Soher, Philip Semanchuk, David Todd, Xiao Ji, Dinesh Deelchand, James Joers, Gulin Oz, and Karl Young. Vespa: integrated applications for rf pulse design, spectral simulation and mrs data analysis.Magnetic resonance in medicine, 90(3):823–838, 2023

    work page 2023

  11. [11]

    Daniel GQ Chong, Roland Kreis, Christine S Bolliger, Chris Boesch, and Johannes Slotboom. Two-dimensional linear-combination model fitting of magnetic resonance spectra to define the macromolecule baseline using fi- taid, a fitting tool for arrays of interrelated datasets.Magnetic Resonance Materials in Physics, Biology and Medicine, 24(3):147–164, 2011

    work page 2011

  12. [12]

    Inspector-magnetic resonance spectroscopy software.Columbia TechVenture (CTV): p

    C Juchem. Inspector-magnetic resonance spectroscopy software.Columbia TechVenture (CTV): p. License CU17130, 2016. 18

    work page 2016

  13. [13]

    Java-based graphical user interface for the mrui quantitation package.Magnetic resonance materials in physics, biology and medicine, 12(2):141–152, 2001

    A Naressi, C Couturier, JM Devos, M Janssen, C Mangeat, R de Beer, and D Graveron-Demilly. Java-based graphical user interface for the mrui quantitation package.Magnetic resonance materials in physics, biology and medicine, 12(2):141–152, 2001

    work page 2001

  14. [14]

    Andrew A Maudsley, C Domenig, V Govind, A Darkazanli, C Studholme, K Arheart, and C Bloomer. Mapping of brain metabolite distributions by volumetric proton mr spectroscopic imaging (mrsi).Magnetic Resonance in Medicine: An Official Journal of the International Society for Magnetic Resonance in Medicine, 61(3):548–559, 2009

    work page 2009

  15. [15]

    Comprehensive processing, display and analysis for in vivo mr spectroscopic imaging.NMR in Biomedicine, 19(4):492–503, 2006

    AA Maudsley, A Darkazanli, JR Alger, LO Hall, N Schuff, C Studholme, Y Yu, A Ebel, A Frew, D Goldgof, et al. Comprehensive processing, display and analysis for in vivo mr spectroscopic imaging.NMR in Biomedicine, 19(4):492–503, 2006

    work page 2006

  16. [16]

    Sivic: open-source, standards-based software for dicom mr spectroscopy workflows.Interna- tional journal of biomedical imaging, 2013(1):169526, 2013

    Jason C Crane, Marram P Olson, and Sarah J Nelson. Sivic: open-source, standards-based software for dicom mr spectroscopy workflows.Interna- tional journal of biomedical imaging, 2013(1):169526, 2013

    work page 2013

  17. [17]

    Contribution of macromolecules to brain 1h mr spectra: experts’ consensus recommendations.NMR in Biomedicine, 34(5):e4393, 2021

    Cristina Cudalbu, Kevin L Behar, Pallab K Bhattacharyya, Wolfgang Bogner, Tamas Borbath, Robin A de Graaf, Rolf Gruetter, Anke Henning, Christoph Juchem, Roland Kreis, et al. Contribution of macromolecules to brain 1h mr spectra: experts’ consensus recommendations.NMR in Biomedicine, 34(5):e4393, 2021

    work page 2021

  18. [18]

    Deep-er: Deep learning eccentric reconstruction for fast high-resolution neurometabolic imaging.NeuroImage, 309:121045, 2025

    Paul J Weiser, Georg Langs, Wolfgang Bogner, Stanislav Motyka, Bernhard Strasser, Polina Golland, Nalini Singh, Jorg Dietrich, Erik Uhlmann, Tracy Batchelor, et al. Deep-er: Deep learning eccentric reconstruction for fast high-resolution neurometabolic imaging.NeuroImage, 309:121045, 2025

    work page 2025

  19. [19]

    Incorporation of a spectral model in a convolu- tional neural network for accelerated spectral fitting.Magnetic resonance in medicine, 81(5):3346–3357, 2019

    Saumya S Gurbani, Sulaiman Sheriff, Andrew A Maudsley, Hyunsuk Shim, and Lee AD Cooper. Incorporation of a spectral model in a convolu- tional neural network for accelerated spectral fitting.Magnetic resonance in medicine, 81(5):3346–3357, 2019

    work page 2019

  20. [20]

    Physics-informed deep learning approach to quantification of human brain metabolites from magnetic resonance spectroscopy data.Computers in Bi- ology and Medicine, 158:106837, 2023

    Amirmohammad Shamaei, Jana Starcukova, and Zenon Starcuk Jr. Physics-informed deep learning approach to quantification of human brain metabolites from magnetic resonance spectroscopy data.Computers in Bi- ology and Medicine, 158:106837, 2023

    work page 2023

  21. [21]

    A deep autoencoder for fast spectral-temporal fitting of dynamic deuterium metabolic imaging data at 7t.NeuroImage, page 121632, 2025

    Aaron Paul Osburg, Amirmohammed Shamaei, Bernhard Strasser, Fabian Niess, Anna Duguid, Viola Bader, Sabina Frese, Lukas Hingerl, Hauke Fis- cher, William T Clarke, et al. A deep autoencoder for fast spectral-temporal fitting of dynamic deuterium metabolic imaging data at 7t.NeuroImage, page 121632, 2025

    work page 2025

  22. [22]

    Tensorfit: A torch-based tool for ultrafast metabolite fitting of large mrsi data sets.Magnetic Resonance in Medicine, 92(2):447–458, 2024

    Federico Turco, Milena Capiglioni, Guodong Weng, and Johannes Slot- boom. Tensorfit: A torch-based tool for ultrafast metabolite fitting of large mrsi data sets.Magnetic Resonance in Medicine, 92(2):447–458, 2024. 19

    work page 2024

  23. [23]

    Quantification of human brain metabolites from in vivo1h nmr magnitude spectra using automated arti- ficial neural network analysis.Journal of Magnetic Resonance, 154(1):1–5, 2002

    Yrj¨ o Hiltunen, Jouni Kaartinen, Juhani Pulkkinen, Anna-Maija H¨ akkinen, Nina Lundbom, and Risto A Kauppinen. Quantification of human brain metabolites from in vivo1h nmr magnitude spectra using automated arti- ficial neural network analysis.Journal of Magnetic Resonance, 154(1):1–5, 2002

    work page 2002

  24. [24]

    Fast quantification of proton magnetic resonance spectroscopic imaging with artificial neural networks.Journal of Magnetic Resonance, 183(1):110–122, 2006

    Himanshu Bhat, Balasrinivasa Rao Sajja, and Ponnada A Narayana. Fast quantification of proton magnetic resonance spectroscopic imaging with artificial neural networks.Journal of Magnetic Resonance, 183(1):110–122, 2006

    work page 2006

  25. [25]

    Magnetic resonance spectroscopy quantification using deep learning

    Nima Hatami, Micha¨ el Sdika, and H´ el` ene Ratiney. Magnetic resonance spectroscopy quantification using deep learning. InInternational Confer- ence on Medical Image Computing and Computer-Assisted Intervention, pages 467–475. Springer, 2018

    work page 2018

  26. [26]

    Intact metabolite spectrum mining by deep learning in proton magnetic resonance spectroscopy of the brain

    Hyeong Hun Lee and Hyeonjin Kim. Intact metabolite spectrum mining by deep learning in proton magnetic resonance spectroscopy of the brain. Magnetic resonance in medicine, 82(1):33–48, 2019

    work page 2019

  27. [27]

    Mrsnet: metabo- lite quantification from edited magnetic resonance spectra with convolu- tional neural networks.arXiv preprint arXiv:1909.03836, 2019

    M Chandler, C Jenkins, SM Shermer, and FC Langbein. Mrsnet: metabo- lite quantification from edited magnetic resonance spectra with convolu- tional neural networks.arXiv preprint arXiv:1909.03836, 2019

    work page arXiv 1909

  28. [28]

    A wavelet scattering convolutional network for magnetic resonance spec- troscopy signal quantitation

    Amirmohammad Shamaei, Jana Starcukov´ a, and Zenon Starcuk Jr. A wavelet scattering convolutional network for magnetic resonance spec- troscopy signal quantitation. InBiosignals, pages 268–275, 2021

    work page 2021

  29. [29]

    Rudy Rizzo, Martyna Dziadosz, Sreenath P Kyathanahally, Amirmoham- mad Shamaei, and Roland Kreis. Quantification of mr spectra by deep learning in an idealized setting: Investigation of forms of input, network architectures, optimization by ensembles of networks, and training bias. Magnetic Resonance in Medicine, 89(5):1707–1727, 2023

    work page 2023

  30. [30]

    Strategies to minimize out-of- distribution effects in data-driven mrs quantification.arXiv preprint arXiv:2511.23135, 2025

    Julian P Merkofer, Antonia Kaiser, Anouk Schrantee, Oliver J Gurney- Champion, and Ruud JG van Sloun. Strategies to minimize out-of- distribution effects in data-driven mrs quantification.arXiv preprint arXiv:2511.23135, 2025

    work page arXiv 2025

  31. [31]

    Deep learning can accelerate and quantify simulated localized correlated spec- troscopy.Scientific Reports, 11(1):8727, 2021

    Zohaib Iqbal, Dan Nguyen, Michael Albert Thomas, and Steve Jiang. Deep learning can accelerate and quantify simulated localized correlated spec- troscopy.Scientific Reports, 11(1):8727, 2021

    work page 2021

  32. [32]

    Dicheng Chen, Meijin Lin, Huiting Liu, Jiayu Li, Yirong Zhou, Taishan Kang, Liangjie Lin, Zhigang Wu, Jiazheng Wang, Jing Li, et al. Mag- netic resonance spectroscopy quantification aided by deep estimations of imperfection factors and macromolecular signal.IEEE Transactions on Biomedical Engineering, 71(6):1841–1852, 2024. 20

    work page 2024

  33. [33]

    Phive: A physics-informed variational encoder enables rapid spectral fitting of brain metabolite mapping at 7t.Medical Image Analysis, page 104014, 2026

    Amirmohammad Shamaei, Amir Buchali, Eva Niess, Lukas Hingerl, Bern- hard Strasser, Aaron Osburg, Korbinian Eckstein, Wolfgang Bogner, and Stanislav Motyka. Phive: A physics-informed variational encoder enables rapid spectral fitting of brain metabolite mapping at 7t.Medical Image Analysis, page 104014, 2026

    work page 2026

  34. [34]

    A ‘self-referential’weight matrix

    J¨ urgen Schmidhuber. A ‘self-referential’weight matrix. InInternational conference on artificial neural networks, pages 446–450. Springer, 1993

    work page 1993

  35. [35]

    HyperNetworks

    David Ha, Andrew Dai, and Quoc V Le. Hypernetworks.arXiv preprint arXiv:1609.09106, 2016

    work page internal anchor Pith review arXiv 2016

  36. [36]

    Hypermorph: Amortized hyperparameter learning for image registration

    Andrew Hoopes, Malte Hoffmann, Bruce Fischl, John Guttag, and Adrian V Dalca. Hypermorph: Amortized hyperparameter learning for image registration. InInternational conference on information processing in medical imaging, pages 3–17. Springer, 2021

    work page 2021

  37. [37]

    Bayesian hypernetworks.arXiv preprint arXiv:1710.04759, 2017

    David Krueger, Chin-Wei Huang, Riashat Islam, Ryan Turner, Alexandre Lacoste, and Aaron Courville. Bayesian hypernetworks.arXiv preprint arXiv:1710.04759, 2017

    work page arXiv 2017

  38. [38]

    Quantifying generative model un- certainty in posterior sampling methods for computational imaging

    Canberk Ekmekci and Mujdat Cetin. Quantifying generative model un- certainty in posterior sampling methods for computational imaging. In Conference on Neural Information Processing Systems (NeurIPS), Deep Inverse Workshop. Curran Associates, 2023

    work page 2023

  39. [39]

    Estimating epistemic and aleatoric uncertainty with a single model.Advances in Neural Infor- mation Processing Systems, 37:109845–109870, 2024

    Matthew Chan, Maria Molina, and Chris Metzler. Estimating epistemic and aleatoric uncertainty with a single model.Advances in Neural Infor- mation Processing Systems, 37:109845–109870, 2024

    work page 2024

  40. [40]

    Sim- ple and scalable predictive uncertainty estimation using deep ensembles

    Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Sim- ple and scalable predictive uncertainty estimation using deep ensembles. Advances in neural information processing systems, 30, 2017

    work page 2017

  41. [41]

    Espirit—an eigenvalue approach to autocalibrating parallel mri: where sense meets grappa.Magnetic resonance in medicine, 71(3):990–1001, 2014

    Martin Uecker, Peng Lai, Mark J Murphy, Patrick Virtue, Michael Elad, John M Pauly, Shreyas S Vasanawala, and Michael Lustig. Espirit—an eigenvalue approach to autocalibrating parallel mri: where sense meets grappa.Magnetic resonance in medicine, 71(3):990–1001, 2014

    work page 2014

  42. [42]

    Paul J Weiser, Georg Langs, Stanislav Motyka, Wolfgang Bogner, S´ ebastien Courvoisier, Malte Hoffmann, Antoine Klauser, and Ovidiu C Andronesi. Walinet: A water and lipid identification convolutional neural network for nuisance signal removal in 1 hˆ 1 h mr spectroscopic imaging.Magnetic Resonance in Medicine, 93(4):1430–1442, 2025

    work page 2025

  43. [43]

    Walinet+: A water and lipid identification neural network for nuisance removal of water-unsuppressed magnetic resonance spectroscopic 21 imaging

    Paul Weiser, Georg Langs, Stanislav Motyka, Wolfgang Bogner, Sebastien Courvoisier, Gulnur Ungan, Malte Hoffmann, Antoine Klauser, and Ovidiu Andronesi. Walinet+: A water and lipid identification neural network for nuisance removal of water-unsuppressed magnetic resonance spectroscopic 21 imaging. InProceedings of the International Society for Magnetic Re...

    work page 2025