Pith. sign in

REVIEW 2 major objections 4 minor 2 cited by

Meta-learning Slice-to-Volume Reconstruction in Fetal Brain MRI using Implicit Neural Representations

T0 review · 2 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A compact two-SIREN architecture performs fetal MRI slice-to-volume reconstruction that withstands severe motion and image corruption without pre-alignment, and a meta-learned initialization cuts reconstruction time by half.

desk verdict Solid subfield improvement in fetal brain SVR with strong simulated-data gains, but the core pose-estimation mechanism is never directly validated. read the letter →

arxiv 2505.09565 v1 pith:GPBG7PRX submitted 2025-05-14 eess.IV cs.AIcs.CV

Maik Dannecker , Thomas Sanchez , Meritxell Bach Cuadra , Özgün Turgut , Anthony N. Price , Lucilio Cordero-Grande , Vanessa Kyriakopoulou , Joseph V. Hajnal
show 1 more author
Daniel Rueckert
This is my paper · ORCID
classification eess.IVcs.AIcs.CV
keywords fetalbrainMRIslice-to-volumereconstructionimplicitneuralrepresentationSIRENmeta-learningmotioncorrectionsuper-resolutionoutlierhandling
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 SSVR, a slice-to-volume reconstruction method for fetal brain MRI built entirely from two implicit neural networks with sinusoidal activations (SIRENs): one estimates each slice's rigid motion and outlier weighting, the other builds a continuous super-resolved 3D volume. The authors claim this compact design removes the need for slice pre-alignment and handles severe inter-slice motion and image artifacts that derail existing methods. They further claim that a fully self-supervised meta-learned initialization, trained on simulated or on as few as ten real subjects, improves reconstruction quality and cuts reconstruction time by about half. If right, motion-corrupted fetal MRI could be reconstructed faster and more reliably across centers without supervised training data.

What carries the argument

The paper's central object is a pair of SIRENs: a Slice Module that takes a per-slice encoding (stack index, slice index) and outputs a rigid transform and outlier weights, and an SR Module that takes homogeneous coordinates and predicts voxel intensity. The Slice Module's two output heads carry motion correction and outlier handling, while the SR Module carries super-resolution through a continuous coordinate-to-intensity mapping. The learning procedure wraps both modules in Reptile-style first-order meta-learning, which produces a shared weight initialization that adapts quickly to new reconstruction tasks. The forward model approximates the slice point spread function by Monte Carlo sampling, with the number of samples increasing quadratically over optimization.

What would settle it

Run the method on simulated slices whose true 3D positions are known, then scramble the slice indices fed into the motion network: if reconstruction quality stays the same, the network is not actually using slice identity to estimate pose; if quality collapses, the identity-to-pose mapping is load-bearing. A cleaner quantitative check is to compare the estimated six-degree-of-freedom transforms with the known simulated transforms and see whether pose error predicts the reported quality gap.

Watch

Extended reading notes

Core claim

The central claim is that motion correction, outlier handling, and super-resolution can be learned jointly by two SIRENs rather than by separate registration or learned pre-alignment. The Slice Module maps the pair (stack index, slice index) to a six-degree-of-freedom rigid transform plus per-slice intensity scaling and weight; the SR Module maps homogeneous 3D coordinates to intensity and is queried during inference. On simulated data with severe corruption, the meta-learned model reports PSNR 20.03 versus 14.28 for NeSVoR and reconstruction time 45 seconds versus 135 seconds. The authors argue that the fully INR-based design lets both modules share a meta-learned initialization, and that self-supervised meta-learning provides an implicit prior that makes the model robust under low-data, cross-center conditions.

Load-bearing premise

The whole scheme rests on one small network being able to figure out each slice's exact 3D position and how much to trust that slice from just the slice's index, with no spatial information and no prior knowledge of the motion; the paper shows this works empirically but does not explain why the network's optimization reliably lands on the correct poses rather than a wrong alignment.

Editorial extensions

If this is right

  • Severe motion corruption no longer requires a separate slice pre-alignment step: on simulated severe cases the meta-learned model reports higher PSNR and SSIM than NeSVoR without any external registration.
  • Reconstruction time drops to roughly one third to one half of NeSVoR's time (45 seconds versus 135 seconds on severe simulated cases), and the four-hour meta-learning cost breaks even after about 49 reconstructions.
  • A meta-initialization trained on simulated data transfers to clinical data from a second center with matching slice resolution, suggesting cross-domain robustness.
  • Training the meta-initialization on just ten real subjects yields substantial gains, making in-house clinical training feasible without large annotated datasets.
  • With fewer input stacks (three instead of six), SSVR retains higher quality than the baselines, indicating robustness when redundant slice information is limited.

Reading between the lines

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

  • If the Slice Module genuinely recovers six-degree-of-freedom poses from slice indices alone, the same two-SIREN design could extend to other slice-based acquisitions such as ultrasound or histology, where per-slice identity encodes acquisition order.
  • A natural testable extension is to compare the estimated transforms against known ground-truth poses on simulated data, mapping which corruption regimes the implicit pose regression handles and which it fails.
  • Pixel-wise rather than slice-wise outlier weighting, which the paper lists as future work, is a plausible improvement when artifacts are localized within otherwise high-quality slices.
  • The quadratic scheduling of PSF samples implies a speed-accuracy trade-off that could be probed: lowering the maximum sample count may reveal how much early low-frequency information shapes the meta-learned initialization.
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

2 major / 4 minor

Summary. The paper proposes SSVR, a slice-to-volume reconstruction method for fetal brain MRI built entirely from two SIRENs: a Slice Module that maps a per-slice encoding (stackidx, sliceidx) to a 6-DoF rigid transform and per-slice intensity scale/weight, and an SR Module that represents the high-resolution volume as an implicit neural function. The two modules are optimized jointly through the self-supervised forward-model loss of Eq. (5), with optional Reptile-based meta-learning that initializes the model weights from simulated or real training cases. The authors evaluate on simulated dHCP data with known ground truth, on 121 real dHCP subjects, and on 40 CHUV subjects, comparing against NeSVoR and SVRTK. The main claims are improved robustness under severe motion and image corruption, faster reconstruction (up to about 50% time reduction), and improved performance with a meta-learned initialization trained on as few as 10 subjects.

Significance. If the central claims hold, the paper offers a practically relevant contribution: a compact, fully INR-based SVR pipeline (about 800k parameters versus 4.8M for NeSVoR) that removes the need for external slice pre-alignment and achieves large, consistent gains under severe simulated motion corruption (Table 1: PSNR 20.03 vs 14.28 for NeSVoR at µ=4). The simulated-data evaluation uses true ground truth and is the strongest evidence in the paper, and the disclosure that the real-data references are generated with SVoRT+NeSVoR is an honest statement of a real limitation. The meta-learning experiments on 10 subjects address an important practical low-data regime. However, the motion-correction claim is not directly verified: the Slice Module's estimated poses are never compared with the known ground-truth transforms on simulated data, and the real-data references are produced by one of the baseline methods, so both the mechanism and the comparative real-world evidence need strengthening.

major comments (2)
  1. [§3.1, Eq. (5), §5.5, Table 1] The paper never evaluates whether the Slice Module actually recovers the correct slice poses. The Slice Module receives only the pair (stackidx, sliceidx) and must regress a 6-DoF rigid transform and outlier weights purely from the joint self-supervised loss of Eq. (5), with no spatial input, no pose supervision, and no registration prior. Yet every reported reconstruction metric in Section 5.5 is computed after registering the reconstruction to the reference volume, so a high PSNR/SSIM does not distinguish genuine slice realignment from a strong implicit brain prior in the SR Module that absorbs pose errors in volume space. The headline severe-motion result in Table 1 (SSVRmeta1 20.03 PSNR vs NeSVoR 14.28) could in principle arise without accurate pose estimation. Because pose regression is the differentiating component of the method, the paper should report, on the simulated data where ground-truth transforms are known, slice-level pose errors (e.g., mean rotation/translation error) or at least alignment residuals; the ablation in Table 5 uses only PSNR/SSIM and does not close this gap.
  2. [§5.4, §6.2, Table 4] The real-data evaluation is partially circular and the manuscript's characterization of it as 'objective' is overstated. Section 5.4 states that references for dHCP and CHUV data are generated by SVoRT followed by NeSVoR, and Section 6.2 acknowledges that NeSVoR's high PSNR in Table 4 is due to the reference being generated by NeSVoR itself. This evaluation design gives NeSVoR a reference that matches its own reconstruction bias while SSVR is measured against a reference produced by a competing method. The stack-ablation study in Section 6.2 is therefore not an 'objective measurement for robustness' as claimed; it is a comparison in which one baseline is advantaged by construction. The authors should either report results against an independently generated reference (e.g., SVRTK or a consensus of multiple methods), or explicitly restrict real-data conclusions to relative robustness trends and frame the absolute PSNR gaps with appropriate caveats.
minor comments (4)
  1. [§5.3] The meta-learning hyperparameters are not reported: Algorithm 1 depends on inner and outer learning rates α and β and the number of inner steps, but Section 5.3 gives only the training time (about 4 hours) and the data composition. Without these values, the meta-learning results are not reproducible and the claimed speed/quality advantages cannot be independently assessed.
  2. [Table 3] Table 3 appears to be an unrelated placeholder: its caption refers to multimodal input, latent codes, Dice scores, and scan/birth age, but the table content is missing and the actual ablation results are presented in Table 5. This needs to be corrected.
  3. [Abstract and §6.1] The abstract states 'up to 50% reduction in reconstruction time,' but Section 6.1 reports a 66% reduction compared to NeSVoR at µ=4 (45 s vs 135 s in Table 1). The claims should be made consistent.
  4. [§6.1, Table 1] Table 1 reports metrics only for µ=1 and µ=4, while the text describes evaluations across five corruption levels µ=1 to 5. The full set of numbers (or a supplementary table) would make the performance-vs-corruption trend easier to verify than relying solely on the curves in Fig. 3b.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reconstruction is a self-supervised inverse problem, and the real-world reference generation is disclosed and favors the baseline rather than the proposed method.

full rationale

The derivation is self-contained. The core objective (Eq. 5) is a data-consistency loss between acquired slices I_i and slices simulated from the SR volume through the Slice Module's estimated rigid transform and outlier weights; no quantity in the loss is pre-fitted from the evaluation targets, and no supervised label enters the optimization. Meta-learning (Algorithm 1, Eqs. 7-9) is a self-supervised Reptile-style initialization on either simulated slices or real acquisition stacks, with evaluation subjects explicitly excluded: "None of the subjects used for evaluation in Section 6 were included in the meta-learning phase." Simulated evaluation uses the original ground-truth volumes that generated the slices, so the metric is external to the model. For real-world data, references are generated by SVoRT+NeSVoR, not by SSVR; the paper itself flags the resulting confound in Sec. 6.2: "NeSVoR's high PSNR is due to the reference being generated by NeSVoR itself (after pre-alignment with SVoRT). Therefore, contrary to the other methods, NeSVoR encounters no domain shift between reconstruction and reference." This favors the baseline and is a disclosed benchmark limitation, not a circular step. The absence of direct pose-accuracy evaluation for the Slice Module (Sec. 3.3.1) is a genuine validation gap, but it does not make any prediction equivalent to its inputs; the method's outputs are still determined by minimizing reconstruction error against the acquired slices. Self-citations (Jiang et al. 2007, Kuklisova-Murgasova et al. 2012, Tourbier et al. 2015, etc.) are used only for standard PSF models and SVR baselines, and are not load-bearing for the novelty claims.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The method's claims rest on a standard MRI forward model, an empirical architectural assumption about the Slice Module, and a meta-learning transfer assumption. The principal free choices are iteration schedules and PSF approximation constants; the reported speedups depend on the hand-picked iteration reduction for meta-learned initializations.

free parameters (4)
  • iteration_scaling_alpha = 125 (standard init), 50 (meta-learned init)
    Section 5.2 sets iteration_max = alpha * total_voxels / batch_size; meta-learned models are run with fewer iterations, which directly produces the reported time reductions. This is a hand-chosen schedule, not a fixed convergence criterion.
  • PSF_MC_samples_K_max = 64
    Section 5.2: K = max(1, 64 * (iteration_current / iteration_max)^2); the quadratic ramp and cap are chosen for speed and are part of the forward-model approximation.
  • PSF_covariance_in_plane_factor = 1.2
    Section 5.2: Sigma = diag((1.2 r_x / 2.355)^2, (1.2 r_y / 2.355)^2, (r_z / 2.355)^2); the 1.2 factor is adopted from prior work (Jiang et al., 2007) but still a modeling choice for the slice profile.
  • meta_learning_inner_and_outer_rates = not reported
    Section 3.4 and 5.3 describe Reptile-based meta-training over 200 cases but do not report the inner/outer learning rates or inner-loop length, leaving a reproducibility gap for the meta-initialization.
assumptions (5)
  • domain assumption The forward model Xi = Bi Ti V + epsilon_i with an anisotropic Gaussian PSF and additive Gaussian noise describes the acquisition.
    Eq. (1)-(3); inherited from Rousseau et al. (2006), Jiang et al. (2007), and Xu et al. (2023). The entire reconstruction loss (Eq. 4-5) is built on this generative model.
  • domain assumption Slice motion is rigid, represented by 3 Euler angles and 3 translations per slice.
    Section 3.3.1; excludes non-rigid motion, through-plane deformation, and intra-slice distortion, which can occur in fetal MRI.
  • ad hoc to paper A SIREN fed only with (stackidx, sliceidx) can regress accurate per-slice rigid transforms and outlier weights.
    Section 3.3; the paper supports this empirically (Tables 1, 2, 5) but provides no analysis of the loss landscape or convergence guarantees; this is the load-bearing mechanism of the proposed architecture.
  • ad hoc to paper Meta-learned weight initializations from 10 simulated dHCP subjects transfer to unseen dHCP and CHUV subjects.
    Section 5.3, 6.2, 6.3; on CHUV data the quality gain over standard initialization is small (PSNR 19.90 vs 19.72 at 3 stacks), so the transfer benefit is only weakly supported, although the speed gain is clear.
  • ad hoc to paper Real-world evaluation references produced by SVoRT + NeSVoR are adequate substitutes for ground truth.
    Section 5.4; this makes NeSVoR comparisons partially circular (acknowledged in Section 6.2) and assumes the SVoRT+NeSVoR pipeline is unbiased for scoring all methods.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Meta-learning Slice-to-Volume Reconstruction in Fetal Brain MRI using Implicit Neural Representations." pith.science (2026). https://pith.science/paper/GPBG7PRX

@misc{pith2026250509565,
  author       = {Pith},
  title        = {Pith review of: Meta-learning Slice-to-Volume Reconstruction in Fetal Brain MRI using Implicit Neural Representations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GPBG7PRX}},
  note         = {Machine review of arXiv:2505.09565}
}
read the original abstract

High-resolution slice-to-volume reconstruction (SVR) from multiple motion-corrupted low-resolution 2D slices constitutes a critical step in image-based diagnostics of moving subjects, such as fetal brain Magnetic Resonance Imaging (MRI). Existing solutions struggle with image artifacts and severe subject motion or require slice pre-alignment to achieve satisfying reconstruction performance. We propose a novel SVR method to enable fast and accurate MRI reconstruction even in cases of severe image and motion corruption. Our approach performs motion correction, outlier handling, and super-resolution reconstruction with all operations being entirely based on implicit neural representations. The model can be initialized with task-specific priors through fully self-supervised meta-learning on either simulated or real-world data. In extensive experiments including over 480 reconstructions of simulated and clinical MRI brain data from different centers, we prove the utility of our method in cases of severe subject motion and image artifacts. Our results demonstrate improvements in reconstruction quality, especially in the presence of severe motion, compared to state-of-the-art methods, and up to 50% reduction in reconstruction time.

Figures

Figures reproduced from arXiv: 2505.09565 by the authors.

Figure 1
Figure 1. Overview 3D reconstruction. a) Acquisition stacks of motion corrupted 2D slices across di [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Real-world raw acquisition stacks from dHCP (top) and [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Evaluation of five simulated datasets with increasing simulated motion- and image corruption from left (µ [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Reconstruction results of four cases of di [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Stack ablation on 40 fetal subjects of the dHCP dataset. a) Leftmost column shows the reference generated as described in Section 5.4. [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: Reconstruction convergence. With standard initialization, our [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. PRIME-SVR: Physics-infoRmed Implicit Multi-Echo Slice-to-Volume Reconstruction for Fetal T2 mapping

    physics.med-ph 2026-07 conditional novelty 6.0 of 10

    A self-supervised, physics-regularized neural reconstruction produces high-resolution fetal brain T2 maps at 0.55 T and 1.5 T from multi-echo MRI, with reduced acquisition time.

  2. Physics-Informed Joint Multi-TE Super-Resolution with Implicit Neural Representation for Robust Fetal T2 Mapping

    cs.CV 2025-08 conditional novelty 5.0 of 10

    Joint multi-echo super-resolution reconstruction with a T2-decay physics regularizer produces robust fetal brain T2 maps at 0.55T, even with one stack per echo.

Reference graph

Works this paper leans on

45 extracted references · 36 canonical work pages · cited by 2 Pith papers

  1. [1]

    , author Wang, G

    author Ebner, M. , author Wang, G. , author Li, W. , author Aertsen, M. , author Patel, P.A. , author Aughwane, R. , author Melbourne, A. , author Doel, T. , author Dymarkowski, S. , author De Coppi, P. , et al., year 2020 . title An automated framework for localization, segmentation and super-resolution reconstruction of fetal brain MRI . journal NeuroIm...

    work page 2020

  2. [2]

    , author Abbeel, P

    author Finn, C. , author Abbeel, P. , author Levine, S. , year 2017 . title Model-agnostic meta-learning for fast adaptation of deep networks , in: booktitle International conference on machine learning , organization PMLR . pp. pages 1126--1135

    2017

  3. [3]

    , author Estroff, J.A

    author Gholipour, A. , author Estroff, J.A. , author Barnewolt, C.E. , author Robertson, R.L. , author Grant, P.E. , author Gagoski, B. , author Warfield, S.K. , author Afacan, O. , author Connolly, S.A. , author Neil, J.J. , et al., year 2014 . title Fetal MRI : a technical update with educational aspirations . journal Concepts in Magnetic Resonance Part...

    work page 2014

  4. [4]

    , author Estroff, J.A

    author Gholipour, A. , author Estroff, J.A. , author Warfield, S.K. , year 2010 . title Robust super-resolution volume reconstruction from slice acquisitions: application to fetal brain MRI . journal IEEE transactions on medical imaging volume 29 , pages 1739--1758

    work page 2010

  5. [5]

    , author Antoniou, A

    author Hospedales, T. , author Antoniou, A. , author Micaelli, P. , author Storkey, A. , year 2022 . title Meta-learning in neural networks: A survey . journal IEEE Transactions on Pattern Analysis and Machine Intelligence volume 44 , pages 5149--5169 . :10.1109/TPAMI.2021.3079209

    arXiv 2022

  6. [6]

    , author Khanal, B

    author Hou, B. , author Khanal, B. , author Alansary, A. , author McDonagh, S. , author Davidson, A. , author Rutherford, M. , author Hajnal, J.V. , author Rueckert, D. , author Glocker, B. , author Kainz, B. , year 2018 a. title 3-d reconstruction in canonical co-ordinate space from arbitrarily oriented 2-d images . journal IEEE transactions on medical i...

    work page 2018

  7. [7]

    , author Miolane, N

    author Hou, B. , author Miolane, N. , author Khanal, B. , author Lee, M.C. , author Alansary, A. , author McDonagh, S. , author Hajnal, J.V. , author Rueckert, D. , author Glocker, B. , author Kainz, B. , year 2018 b. title Computing cnn loss and gradients for pose estimation with riemannian geometry , in: booktitle Medical Image Computing and Computer As...

    work page 2018

  8. [8]

    , author Xue, H

    author Jiang, S. , author Xue, H. , author Glover, A. , author Rutherford, M. , author Rueckert, D. , author Hajnal, J.V. , year 2007 . title MRI of moving subjects using multislice snapshot images with volume reconstruction ( SVR ): application to fetal, neonatal, and adult brain studies . journal IEEE transactions on medical imaging volume 26 , pages 967--980

    work page 2007

Show all 45 references
  1. [9]

    , author Steinberger, M

    author Kainz, B. , author Steinberger, M. , author Wein, W. , author Kuklisova-Murgasova, M. , author Malamateniou, C. , author Keraudren, K. , author Torsney-Weir, T. , author Rutherford, M. , author Aljabar, P. , author Hajnal, J.V. , et al., year 2015 . title Fast volume re...

    2015

  2. [10]

    , author Ba, J

    author Kingma, D.P. , author Ba, J. , year 2014 . title Adam: A method for stochastic optimization . journal arXiv preprint arXiv:1412.6980

    2014 arXiv

  3. [11]

    , author Quaghebeur, G

    author Kuklisova-Murgasova, M. , author Quaghebeur, G. , author Rutherford, M.A. , author Hajnal, J.V. , author Schnabel, J.A. , year 2012 . title Reconstruction of fetal brain MRI with intensity matching and complete outlier removal . journal Medical image analysis volume 16 ...

    2012

  4. [12]

    , author Srinivasan, P.P

    author Mildenhall, B. , author Srinivasan, P.P. , author Tancik, M. , author Barron, J.T. , author Ramamoorthi, R. , author Ng, R. , year 2021 . title Nerf: Representing scenes as neural radiance fields for view synthesis . journal Communications of the ACM volume 65 , pages 99--106

    2021

  5. [13]

    , author Evans, A

    author M\"uller, T. , author Evans, A. , author Schied, C. , author Keller, A. , year 2022 . title Instant neural graphics primitives with a multiresolution hash encoding . journal ACM Trans. Graph. volume 41 , pages 102:1--102:15 . :10.1145/3528223.3530127

    2022

  6. [14]

    , author Achiam, J

    author Nichol, A. , author Achiam, J. , author Schulman, J. , year 2018 . title On first-order meta-learning algorithms . journal arXiv preprint arXiv:1803.02999

    2018 arXiv

  7. [15]

    , author Florence, P

    author Park, J.J. , author Florence, P. , author Straub, J. , author Newcombe, R. , author Lovegrove, S. , year 2019 . title Deepsdf: Learning continuous signed distance functions for shape representation , in: booktitle Proceedings of the IEEE/CVF conference on computer visio...

    2019

  8. [16]

    , author Wang, L

    author Pei, Y. , author Wang, L. , author Zhao, F. , author Zhong, T. , author Liao, L. , author Shen, D. , author Li, G. , year 2020 . title Anatomy-guided convolutional neural network for motion correction in fetal brain mri , in: booktitle Machine Learning in Medical Imagin...

    2020

  9. [17]

    , author Sparks, R

    author P \'e rez-Garc \'i a, F. , author Sparks, R. , author Ourselin, S. , year 2021 . title TorchIO : a python library for efficient loading, preprocessing, augmentation and patch-based sampling of medical images in deep learning . journal Computer Methods and Programs in Bi...

    2021

  10. [18]

    , author Brugger, P.C

    author Prayer, D. , author Brugger, P.C. , author Prayer, L. , year 2004 . title Fetal MRI : techniques and protocols . journal Pediatric radiology volume 34 , pages 685--693

    2004

  11. [19]

    , author Cordero-Grande, L

    author Price, A.N. , author Cordero-Grande, L. , author Hughes, E. , author Hiscocks, S. , author Green, E. , author McCabe, L. , author Hutter, J. , author Ferrazzi, G. , author Deprez, M. , author Roberts, T. , et al., year 2019 . title The developing human connectome projec...

    2019

  12. [20]

    , author Fidon, L

    author Ranzini, M. , author Fidon, L. , author Ourselin, S. , author Modat, M. , author Vercauteren, T. , year 2021 . title MONAI fbs: MONAI -based fetal brain MRI deep learning segmentation

    2021

  13. [21]

    , author Glenn, O.A

    author Rousseau, F. , author Glenn, O.A. , author Iordanova, B. , author Rodriguez-Carranza, C. , author Vigneron, D.B. , author Barkovich, J.A. , author Studholme, C. , year 2006 . title Registration-based approach for reconstruction of high-resolution in utero fetal MR brain...

    2006

  14. [22]

    , year 2014

    author Saleem, S.N. , year 2014 . title Fetal MRI : An approach to practice: A review . journal Journal of advanced research volume 5 , pages 507--523

    2014

  15. [23]

    , author Khan, S

    author Salehi, S.S.M. , author Khan, S. , author Erdogmus, D. , author Gholipour, A. , year 2018 . title Real-time deep pose estimation with geodesic loss for image-to-template rigid registration . journal IEEE transactions on medical imaging volume 38 , pages 470--481

    2018

  16. [24]

    , author Salehi, S.S.M

    author Singh, A. , author Salehi, S.S.M. , author Gholipour, A. , year 2020 . title Deep predictive motion tracking in magnetic resonance imaging: application to fetal imaging . journal IEEE transactions on medical imaging volume 39 , pages 3523--3534

    2020

  17. [25]

    , author Chan, E

    author Sitzmann, V. , author Chan, E. , author Tucker, R. , author Snavely, N. , author Wetzstein, G. , year 2020 a. title Metasdf: Meta-learning signed distance functions . journal Advances in Neural Information Processing Systems volume 33 , pages 10136--10147

    2020

  18. [26]

    , author Martel, J

    author Sitzmann, V. , author Martel, J. , author Bergman, A. , author Lindell, D. , author Wetzstein, G. , year 2020 b. title Implicit neural representations with periodic activation functions . journal Advances in neural information processing systems volume 33 , pages 7462--7473

    2020

  19. [27]

    , author Zollh \"o fer, M

    author Sitzmann, V. , author Zollh \"o fer, M. , author Wetzstein, G. , year 2019 . title Scene representation networks: Continuous 3d-structure-aware neural scene representations , in: booktitle Advances in Neural Information Processing Systems

    2019

  20. [28]

    , author McGinnis, J

    author Stolt-Ans \'o , N. , author McGinnis, J. , author Pan, J. , author Hammernik, K. , author Rueckert, D. , year 2023 . title Nisf: Neural implicit segmentation functions , in: booktitle International Conference on Medical Image Computing and Computer-Assisted Intervention...

    2023

  21. [29]

    , author Hill, D

    author Studholme, C. , author Hill, D. , author Hawkes, D. , year 1999 . title An overlap invariant entropy measure of 3d medical image alignment . journal Pattern Recognition volume 32 , pages 71--86 . :https://doi.org/10.1016/S0031-3203(98)00091-0

    1999 doi

  22. [30]

    , author Mildenhall, B

    author Tancik, M. , author Mildenhall, B. , author Wang, T. , author Schmidt, D. , author Srinivasan, P.P. , author Barron, J.T. , author Ng, R. , year 2021 . title Learned initializations for optimizing coordinate-based neural representations , in: booktitle Proceedings of th...

    2021

  23. [31]

    , author Srinivasan, P

    author Tancik, M. , author Srinivasan, P. , author Mildenhall, B. , author Fridovich-Keil, S. , author Raghavan, N. , author Singhal, U. , author Ramamoorthi, R. , author Barron, J. , author Ng, R. , year 2020 . title Fourier features let networks learn high frequency function...

    2020

  24. [32]

    , author Bresson, X

    author Tourbier, S. , author Bresson, X. , author Hagmann, P. , author Thiran, J.P. , author Meuli, R. , author Cuadra, M.B. , year 2015 . title An efficient total variation algorithm for super-resolution in fetal brain MRI with adaptive regularization . journal NeuroImage vol...

    2015

  25. [33]

    , author Avants, B.B

    author Tustison, N.J. , author Avants, B.B. , author Cook, P.A. , author Zheng, Y. , author Egan, A. , author Yushkevich, P.A. , author Gee, J.C. , year 2010 . title N4ITK : improved N3 bias correction . journal IEEE transactions on medical imaging volume 29 , pages 1310--1320

    2010

  26. [34]

    , author Zhang, T

    author Uus, A. , author Zhang, T. , author Jackson, L.H. , author Roberts, T.A. , author Rutherford, M.A. , author Hajnal, J.V. , author Deprez, M. , year 2020 . title Deformable slice-to-volume registration for motion correction of fetal body and placenta mri . journal IEEE t...

    2020

  27. [35]

    , author Egloff Collado, A

    author Uus, A.U. , author Egloff Collado, A. , author Roberts, T.A. , author Hajnal, J.V. , author Rutherford, M.A. , author Deprez, M. , year 2023 a. title Retrospective motion correction in foetal MRI for clinical applications: existing methods, applications and integration ...

    2023

  28. [36]

    , author Hall, M

    author Uus, A.U. , author Hall, M. , author Payette, K. , author Hajnal, J.V. , author Deprez, M. , author Rutherford, M.A. , author Hutter, J. , author Story, L. , year 2023 b. title Combined quantitative t2* map and structural t2-weighted tissue-specific analysis for fetal b...

    2023

  29. [37]

    , author Bovik, A

    author Wang, Z. , author Bovik, A. , author Sheikh, H. , author Simoncelli, E. , year 2004 . title Image quality assessment: from error visibility to structural similarity . journal IEEE Transactions on Image Processing volume 13 , pages 600--612 . :10.1109/TIP.2003.819861

    2004

  30. [38]

    , author Li, Y

    author Wu, Q. , author Li, Y. , author Xu, L. , author Feng, R. , author Wei, H. , author Yang, Q. , author Yu, B. , author Liu, X. , author Yu, J. , author Zhang, Y. , year 2021 . title IREM : High-resolution magnetic resonance image reconstruction via implicit neural represe...

    2021

  31. [39]

    , author Moyer, D

    author Xu, J. , author Moyer, D. , author Gagoski, B. , author Iglesias, J.E. , author Grant, P.E. , author Golland, P. , author Adalsteinsson, E. , year 2023 . title NeSVoR : implicit neural representation for slice-to-volume reconstruction in MRI . journal IEEE transactions ...

    2023

  32. [40]

    , author Moyer, D

    author Xu, J. , author Moyer, D. , author Grant, P.E. , author Golland, P. , author Iglesias, J.E. , author Adalsteinsson, E. , year 2022 . title SVoRT : Iterative transformer for slice-to-volume registration in fetal brain MRI , in: booktitle International Conference on Medic...

    2022

  33. [41]

    , author Aliasi, M

    author Yeung, P.H. , author Aliasi, M. , author Papageorghiou, A.T. , author Haak, M. , author Xie, W. , author Namburete, A.I. , year 2021 . title Learning to map 2d ultrasound images into 3d space with minimal human annotation . journal Medical Image Analysis volume 70 , pag...

    2021

  34. [42]

    , author Balbastre, Y

    author Young, S.I. , author Balbastre, Y. , author Fischl, B. , author Golland, P. , author Iglesias, J.E. , year 2024 . title Fully convolutional slice-to-volume reconstruction for single-stack mri , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and ...

    2024

  35. [43]

    , author Ortiz-Jim\'enez, G

    author Y\"uce, G. , author Ortiz-Jim\'enez, G. , author Besbinar, B. , author Frossard, P. , year 2022 . title A structured dictionary perspective on implicit neural representations , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...

    2022

  36. [44]

    , author Piven, J

    author Yushkevich, P.A. , author Piven, J. , author Cody Hazlett, H. , author Gimpel Smith, R. , author Ho, S. , author Gee, J.C. , author Gerig, G. , year 2006 . title User-guided 3D active contour segmentation of anatomical structures: Significantly improved efficiency and r...

    2006

  37. [45]

    , author Wang, X

    author Zhang, L. , author Wang, X. , author Rawson, M. , author Balan, R. , author Herskovits, E.H. , author Melhem, E.R. , author Chang, L. , author Wang, Z. , author Ernst, T. , year 2024 . title Motion correction for brain mri using deep learning and a novel hybrid loss fun...

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.