Pith. sign in

REVIEW 4 major objections 5 minor 49 references

Neural Image Unfolding: Flattening Sparse Anatomical Structures using Neural Fields

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A neural field fitted to sparse 3D anatomical points can flatten them into a 2D overview with lower peak distortion than mesh-based unfolding baselines.

desk verdict Novel neural-field unfolding method with real clinical potential, but the headline distortion advantage is confounded by a fidelity trade-off and needs a matched-fidelity comparison before the claim holds. read the letter →

arxiv 2411.18415 v1 pith:BUVYEFUS submitted 2024-11-27 cs.CV

Leonhard Rist , Pluvio Stephan , Noah Maul , Linda Vorberg , Hendrik Ditt , Michael Sühling , Andreas Maier , Bernhard Egger
show 1 more author
Oliver Taubmann
This is my paper · ORCID
classification cs.CV
keywords neuralfieldsimageunfoldingsparsestructuresdistortionregularizationmedicalvisualizationtomographicimagingpointcloudparameterizationdeformationfield
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 versatile, modality-independent method to unfold thin, sparse anatomical structures—such as cerebral vessels, hand bones, the pancreas with adjacent arteries, and the rib cage—into a single 2D overview image. Rather than requiring a densely sampled surface or a global ordering of points, it fits a small neural field that deforms an initial plane so that the plane intersects the sparse target points with minimal local distortion. The paper reports that this approach outperforms mesh-based baselines on maximum distortion for complex sparse structures, and that its multi-scale pairwise distortion regularizer produces smoother, fold-free transformations than Jacobian-based regularizers from neural field registration. If correct, the method gives clinicians a general tool for producing standardized, low-distortion views of structures that currently lack a versatile unfolding technique.

What carries the argument

The central object is a deformation neural field: a three-layer MLP with Leaky ReLU activations and a low-frequency trigonometric embedding that maps points on an initial PCA-derived plane to positions in the tomographic volume. The load-bearing loss is the multi-scale distortion regularizer, which samples point pairs at distances spanning 0.5 to 40 mm and penalizes deviations between their 3D distance and their fixed 2D pixel distance, preventing folds and distributing inevitable distortion away from the target when weighted by an importance map derived from a Euclidean distance transform. A Chamfer-style target loss pulls the deformed plane onto the sparse point set, and an optional intensity-based loss optimizes appearance in the unfolded image.

What would settle it

Fit the model to a sparse point set sampled from a high-curvature 3D helix with radius comparable to the pixel spacing and the same loss weights and normalization used in the paper; if the median distance from target points to the fitted sheet exceeds the 0.5 mm pixel spacing, the single-smooth-manifold assumption—and with it the claimed versatility on complex sparse structures—fails.

Watch

Extended reading notes

Core claim

The central claim is that a pointwise-fitted neural field can serve as a general-purpose unfolding parametrization for sparse anatomical structures: starting from a PCA-aligned planar initialization, a low-frequency multi-layer perceptron learns a deformation field that pulls the plane onto the target centerline points while a multi-scale distance-preserving loss keeps the read-out image geometrically faithful. Alongside this geometric fit, an intensity-based loss recovers non-annotated vessels or maximizes the display of an auxiliary organ mask. The paper demonstrates on four CT applications that this framework yields lower maximum distortions than CeVasMap and ARAP, and that the proposed distortion regularizer leads to smoother results than the Jacobian regularizers used in prior neural field registration work.

Load-bearing premise

The target point set must lie close to a single smooth 2D manifold that a low-frequency neural field can parameterize with acceptable distortion; if the structure is thick, tightly folded, or consists of distant parallel sheets, the unfolded image will miss parts of it.

Editorial extensions

If this is right

  • A single framework can unfold sparse structures such as vascular trees, ducts, or bone systems without requiring global point ordering, removing a limitation of curved planar reformation.
  • The multi-scale pairwise distortion regularizer is a reusable component for any neural-field geometric task that needs smooth, fold-free deformations, independent of medical imaging.
  • The same default configuration works across four anatomies, so new unfolding tasks can start from a known-good setting and only adjust the normalization constant tied to target extent.
  • Combining geometric and intensity losses lets the unfolded image recover structures with missing annotations and display auxiliary targets, such as an organ mask, alongside the primary structure.

Reading between the lines

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

  • Because the neural field is resolution-independent, the same fitted model could be re-evaluated at arbitrary output resolutions, enabling zoomable or progressive unfolding without retraining.
  • The importance map could also gate the image-based loss, focusing appearance optimization near the target or in peripheral areas depending on clinical need, a step the paper mentions but does not explore.
  • For structures that are genuinely volumetric rather than sheet-like, the single-manifold assumption will break; fitting multiple coupled neural patches or adding a thickness-aware cost would be a natural next step.
  • The reported trade-off of slightly larger target-to-sheet distances than mesh baselines suggests hybrid pipelines could use the neural field for coarse, low-distortion unfolding followed by a local refinement to pin exact landmarks.
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. The paper proposes a neural field-based framework for unfolding sparse anatomical structures (e.g., vessel centerlines, bone centerlines) into a 2D image. A low-frequency MLP predicts a displacement field on top of a PCA-initialized plane, and is trained with a Chamfer-like target-distance loss, a multi-scale pairwise distortion loss, and an optional image-based loss. The authors demonstrate the approach on cerebral vessels, hand bones, pancreas with vessels, and rib cages, and claim that it outperforms the mesh-based baselines CeVasMap and ARAP with respect to peak distortion, and that the proposed multi-scale distortion regularizer produces smoother transformations than Jacobian-based neural field regularizers.

Significance. If the quantitative claims were robustly supported, this would be a useful contribution to medical image unfolding: the method is modality-independent, does not require a global point ordering, and integrates geometric and appearance objectives in a single optimization. The multi-scale pairwise distortion regularizer and the importance-sampling scheme are practical ideas that could benefit future work. The paper is also transparent about its experimental setup, including a stopping criterion and ablation details. However, the central comparative claims are currently weakened by evaluation issues that need to be addressed before the paper can be accepted.

major comments (4)
  1. [Sec. 6.1, Sup. Table 2] The headline claim of outperforming mesh-based baselines with respect to peak distortion is confounded by a large fidelity gap. The proposed method reports mean target-to-mesh distances of 0.253-0.651 mm across the four tasks, whereas ARAP and CeVasMap achieve 0.024-0.138 mm. For the rib cage, the proposed method has mean distance 0.651 mm versus 0.035 mm (ARAP) and 0.024 mm (CeVasMap), i.e., above the 0.5 mm pixel spacing the authors themselves use as a threshold. Because Eq. (1) is a weighted sum of target distance and distortion, the lower peak distortion can be partly purchased by leaving targets farther from the fitted manifold. A matched-fidelity comparison (e.g., constraining target distance to a common budget) or a joint distortion-fidelity cost is needed to support the claim of superior distortion.
  2. [Sec. 5.2, Sec. 6] The evaluation of distortion is partly circular: the distortion values reported in Fig. 5 and Sup. Table 2 are computed with Eq. (4), which is exactly the loss term Ld optimized during fitting, and they are measured in a 1 cm radius around the unfolded target, i.e., the region that the importance weighting (Sec. 5.2) explicitly emphasizes via ws in Eq. (4). The baselines do not optimize this loss, so the comparison is not on a neutral common yardstick. Please report an independent metric (e.g., manual assessment by clinicians, landmark-based accuracy, or area distortion measured against a ground-truth mesh) and analyze the sensitivity of the conclusions to the evaluation window size.
  3. [Sec. 6.2, Sup. Table 3] The claim that the proposed multi-scale distortion regularizer yields 'smoother transformations compared to Jacobian formulations' is not supported by the quantitative results in Sup. Table 3. The Jacobian regularizers J1 and J2 achieve much lower maximum distortion (0.957 mm and 1.384 mm, respectively) than the proposed multi-scale baseline (6.057 mm). The authors argue that J1 changes the initial plane structure and J2 stretches vessel radii, but those observations are qualitative. Please provide quantitative morphology-preservation metrics (e.g., fold counts, vessel-radius error, or landmark displacement) to substantiate the superiority of the proposed regularizer over the Jacobian-based alternatives.
  4. [Sec. 5.1, Sec. 7.4] The paper's claim of a 'versatile framework' is only partially supported. The PCA initialization and low-frequency embedding assume that the target point set lies near a single smooth 2-manifold; this assumption is central to the method. The experiments show its limits: the hand-bone case misses 'the last part of the tips,' and the rib-cage target distance remains above the pixel spacing. The paper would be strengthened by an explicit statement of the conditions under which the manifold assumption is violated, and by a diagnostic (e.g., PCA residual or target-distance statistics) that could tell a user whether the method is applicable to a new structure.
minor comments (5)
  1. [Fig. 5] The boxplot uses a logarithmic scale, but the horizontal line at the pixel spacing (0.5 mm) is near the lower end and may be visually misleading; please clarify the axis or use a linear scale for a subset of the data.
  2. [Sec. 5.2] The importance map formula VE = (|min(e(I)-α,0)|+β)/(α+β) is under-specified: please define the units of e(I) (Euclidean distance transform) and state whether α and β are in millimeters or voxels.
  3. [Sec. 4.2] In Eq. (4), the loss uses ||us,1 - us,2||, whereas the text earlier defines Δ as the distance between neighboring pixels; please clarify how the random sampling distances are chosen and how the loss scale relates to the image resolution.
  4. [Sup. Table 3] The column headers 'Ours 10 mm' and 'Ours 50 mm' are ambiguous when read in isolation; consider renaming them to 'Small-scale (10 mm)' and 'Large-scale (50 mm)' to match the terminology in Sec. 6.2.
  5. [Sec. 6.3] The references to Sup. Eq. (7) and Sup. Eq. (8) are clear in the supplementary, but the main text would benefit from a brief description of the sink-like function used for vessel intensity retrieval.

Circularity Check

3 steps flagged · score 6.0 of 10

The headline distortion and target-distance metrics are the exact loss terms optimized in Eq. (1), so the reported peak-distortion advantage is partly a convergence check; the external baseline comparison provides partial independence.

  1. fitted input called prediction [Sec. 4.2, Eq. (4); Sec. 6, Experiments; Sup. Table 2]
    "Ld = 1/S \sum_s ws · (||\hat x_{s,1} − \hat x_{s,2}|| − ||u_{s,1} − u_{s,2}||)^2 , (4) ... Distortion values based on Eq. (4) are reported in the relevant area around the unfolded target (1 cm radius) to be independent of the mesh size."

    The paper's objective is L = wtLt + wdLd + wimLim (Eq. 1), so the neural field is trained to minimize exactly this pairwise distance-distortion term. The quantitative evidence for the claim that the method 'outperforms mesh-based baselines for sparse structures w.r.t. peak distortion' is then reported as the value of Eq. (4). For the proposed method, low Eq. (4) distortion is therefore in part a measure of convergence of the training loss rather than an independent prediction. The comparison is not fully forced because the baselines are not trained with this loss, but the headline metric is the optimized objective itself.

  2. fitted input called prediction [Sec. 4.1, Eq. (2); Sec. 6.1; Sup. Table 2]
    "Lt = 1/N \sum_n min_{s in S} ||\hat x_s − t_n||^2 , (2) ... When investigating target distances, see Sup. Tab. 2, CeVasMap and ARAP show smaller mean values than our neural field by enforcing the target intersection."

    The reported 'Distance' metrics are exactly the Chamfer-like target-to-mesh distance minimized as Lt in Eq. (2). The paper acknowledges that the baselines enforce target intersection and achieve smaller distances, which confirms that this evaluation column is the same fidelity term the network was trained to trade off against distortion. Reporting it as an outcome shows how one training loss was exchanged for another, not an independent measure of the method's utility.

1 more flagged steps
  1. other [Sec. 5.2 Importance Sampling & Map; Sec. 6 Experiments]
    "This emphasizes small distortions close to the target and can help to move inevitable distortions to less important areas. ... Distortion values based on Eq. (4) are reported in the relevant area around the unfolded target (1 cm radius) to be independent of the mesh size."

    The importance map heavily weights the distortion loss Ld near the target and explicitly pushes distortion into the periphery. The evaluation then reports distortion only within a 1 cm radius around the target, i.e. in the spatial region where the model was trained to keep distortion small, while excluding the periphery where the model was trained to absorb distortion. This alignment between the training weighting and the evaluation window makes the reported peak-distortion value partly a consequence of the chosen evaluation region, although the same window is applied to all methods.

full rationale

The paper's central algorithmic contribution is a neural field trained with L = wtLt + wdLd + wimLim (Eq. 1), where Lt is a Chamfer-like target distance (Eq. 2) and Ld is a pairwise distance-distortion term (Eq. 4). The quantitative evidence for the headline claims is Sup. Table 2, whose 'Distortion' numbers are explicitly stated to be based on Eq. (4) and whose 'Distance' numbers are the same closest-point distances as Eq. (2). Thus the evaluation metrics are the training objectives themselves; the reported low peak distortion for the proposed method is in part a convergence check of the fitted loss, not an independent measurement. The external baselines (ARAP and CeVasMap) are not optimized for Eq. (4), so the comparison is not fully forced: the proposed method could in principle lose on this metric. However, the 1 cm evaluation window also coincides spatially with the importance-weighted region where Eq. (4) is emphasized (Sec. 5.2), so the headline metric excludes the periphery in which distortions are deliberately concentrated. No load-bearing self-citation circularity was found: the use of the authors' previous CeVasMap initial plane (Sec. 5.1) is an implementation choice rather than a uniqueness argument, and the cited labeling papers are data-preparation references. The main circularity is therefore the identity between evaluation metric and training loss, which is partial rather than total, supporting a score of 6.

Assumptions & free parameters 11 free parameters · 6 assumptions · 0 invented entities

The central method rests on standard optimization, supervised by hand-chosen weights and a per-anatomy normalization constant c; no new physical entities or external constraints are introduced. The main assumptions concern the manifold structure of the target and the sufficiency of a low-frequency neural field.

free parameters (11)
  • Loss weight w_t = 2
    Set by hand for all experiments to balance target attraction against distortion (Sec 6.1).
  • Loss weight w_d = 1
    Set by hand; relative strength of isometry regularizer (Sec 6.1).
  • Loss weight w_im = 0 for main tasks; 1e-3 for image-based tasks
    Image-based loss weight chosen so appearance does not disrupt geometry (Sec 6.3).
  • Normalization constant c = 750 mm for rib cage; 100 mm for other anatomies
    Hand-set per anatomy for the trigonometric embedding; authors state it scales roughly linearly with target extent (Sec 7.2).
  • Embedding frequency count = 3
    Selected because neighboring displacements are smooth; authors report it 'proved sufficient' without a formal criterion (Sec 5.3).
  • Importance map parameters alpha, beta = alpha=30, beta=0.1 (weak); alpha=10, beta=0.1 (strict)
    Hand-set to control how strongly target-proximal points are weighted during sampling and distortion weighting (Sec 5.2, Sec 6.4).
  • Multi-scale distortion distance range = 0.5 to 40 mm; 10 mm and 50 mm in ablations
    Chosen empirically to combine local isometry and fold prevention; no automatic scale selection (Sec 6.2).
  • Neural network capacity = 3 hidden layers, width 128, LeakyReLU
    Fixed architecture; no reported sensitivity analysis (Sec 5).
  • Initial plane margin = 2x target extent
    Chosen to cushion deformations in peripheral image areas (Sec 7.1).
  • Training budget = 50k point pairs per epoch, 5000 epochs
    Set for comparability; authors note early stopping can shorten it (Sec 5.2, Sec 6.1).
  • Evaluation radius around target = 1 cm
    The reported distortion values are computed in this window rather than globally; the choice favors the method because the importance weighting concentrates distortion outside the window (Sec 6).
assumptions (6)
  • domain assumption Target structures lie on a 2D manifold that admits a low-distortion planar parameterization.
    The whole formulation (Sec 3) models unfolding as a mapping from a 2D image plane into 3D, so non-manifold or highly branched point sets cannot be represented without error.
  • domain assumption The input annotation T is an accurate, sufficiently complete representation of the anatomy of interest.
    The target loss Lt forces the sheet to the annotated points; the limitations section notes annotation errors propagate and can fake stenosis or missing vessels (Sec 7.4).
  • ad hoc to paper A low-frequency neural field with 3-frequency embedding can represent the required deformations.
    The paper argues smooth displacement fields imply a low-frequency bias, but the capacity claim is supported only by experiments, not a bound (Sec 5.3).
  • domain assumption The PCA plane is a sufficient initialization for the deformation optimization.
    The plane gives the initial manifold from which f deforms; if the projection collapses distinct structures, the optimizer must resolve them, and results show this fails for some hand configurations (Sec 5.1, Sec 6.1).
  • domain assumption Euclidean distances in 3D are the right proxy for image distortion.
    Equation (3) describes distortion as deviation of straight-line 3D distance from pixel spacing; this ignores curvature of the structure itself, which is a modeling choice.
  • ad hoc to paper The multi-scale pairwise loss with 0.5-40 mm scales prevents folds across all tested anatomies.
    The authors show small and large scales alone produce artifacts or folds, and the multi-scale choice mitigates them, but this is an empirical observation, not a theorem (Sec 6.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neural Image Unfolding: Flattening Sparse Anatomical Structures using Neural Fields." pith.science (2026). https://pith.science/paper/BUVYEFUS

@misc{pith2026241118415,
  author       = {Pith},
  title        = {Pith review of: Neural Image Unfolding: Flattening Sparse Anatomical Structures using Neural Fields},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BUVYEFUS}},
  note         = {Machine review of arXiv:2411.18415}
}
read the original abstract

Tomographic imaging reveals internal structures of 3D objects and is crucial for medical diagnoses. Visualizing the morphology and appearance of non-planar sparse anatomical structures that extend over multiple 2D slices in tomographic volumes is inherently difficult but valuable for decision-making and reporting. Hence, various organ-specific unfolding techniques exist to map their densely sampled 3D surfaces to a distortion-minimized 2D representation. However, there is no versatile framework to flatten complex sparse structures including vascular, duct or bone systems. We deploy a neural field to fit the transformation of the anatomy of interest to a 2D overview image. We further propose distortion regularization strategies and combine geometric with intensity-based loss formulations to also display non-annotated and auxiliary targets. In addition to improved versatility, our unfolding technique outperforms mesh-based baselines for sparse structures w.r.t. peak distortion and our regularization scheme yields smoother transformations compared to Jacobian formulations from neural field-based image registration.

Figures

Figures reproduced from arXiv: 2411.18415 by the authors.

Figure 1
Figure 1. Unfolding sparse cerebral vessels using a neural field [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Neural unfolding pipeline for the fitting process using randomly sampled points (top) and for inference using a grid input leading to a structured image after read-out (bottom). parametrization whereas the Flatten-Anything-Model [43] uses multiple sub-networks to model the whole mesh pa￾rameterization process for densely sampled surfaces. Im￾plicit neural representations have also found medical appli￾cations [19], e… view at source ↗
Figure 3
Figure 3. Importance sampling (left) and map overlay (right) from an importance definition in volumetric space which is in￾versely proportional to the Euclidean distance to the target (green). incorporation of image appearance can now be readily ex￾tended to the (additional) optimization of other downstream tasks to favor specific appearance properties that may be described analytically or by proximity of embeddings of a suit… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: The resulting surfaces do not exhibit folding, are smooth and follow the object centerlines, see metrics in Sup. Tab. 2. All models are fitted using the same config￾5 [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Logarithmic distortion distribution as boxplot (me￾dian line, std-dev. boxes, maximum whiskers) within 1 cm radius around unfolded target for each anatomy and method. Horizontal line of pixel spacing δ, approx. indicating the beginning of severe distortions. Red dots r…
Figure 6
Figure 6. Figure 6: Distortion regularizers image and mesh results for cerebral vessels. Two Jacobian regularizers (top) [5, 37] vs. our neighborhood regularization (bottom) with small, large and multi￾scale (0.5-40 mm) distances. Artifacts are highlighted in yellow. in [PITH_FULL_IMAGE:…
Figure 7
Figure 7. Figure 7: Use cases of the image-based loss. In the top row, we apply Li to the auxiliary task of displaying the pancreatic or￾gan mask (blue) when unfolding pancreatic vessels. In the bot￾tom row, we unfold the cerebral vasculature, removing (basilar and vertebral) artery segme…
Figure 9
Figure 9. Figure 9: Influence of sampling schemes (in top right corners) in an early fitting stage (500 epochs) when using only 500 samples. Comparing unfinished unfoldings with uniform random (left), hy￾brid (center) [34] and our importance sampling (right). target points (rib tips, vess…
Figure 10
Figure 10. Figure 10: Distortion examples when the distance between neigh￾boring read-out points does not match the image spacing. Left: volumetric readout space with a rendered and cropped head CT in axial direction. Read-out points are displayed in red showing ev￾ery 8th image row and co…
Figure 11
Figure 11. Figure 11: Logarithmic distortion distribution as boxplot (median line, std-dev. boxes, maximum whiskers) within 1 cm radius around unfolded target for each case per anatomy and method. Horizontal line (dotted) of pixel spacing δ, approx. indicates the beginning of severe distor…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

49 extracted references · 48 canonical work pages

  1. [1]

    V olu- metric Parameterization of the Placenta to a Flattened Tem- plate

    S Mazdak Abulnaga, Esra Abaci Turk, Mikhail Bessmeltsev, P Ellen Grant, Justin Solomon, and Polina Golland. V olu- metric Parameterization of the Placenta to a Flattened Tem- plate. IEEE Trans. Med. Imag., 41:925–936, 2022. 2

    work page 2022

  2. [2]

    Intrinsic parameterizations of surface meshes

    Mathieu Desbrun, Mark Meyer, and Pierre Alliez. Intrinsic parameterizations of surface meshes. In Computer graphics forum, pages 209–218, 2002. 2

    work page 2002

  3. [3]

    Richter, Timur Bagautdinov, Pierre Baque, and Pascal Fua

    Beno ˆıt Guillard, Edoardo Remelli, Artem Lukoianov, Pierre Yvernay, Stephan R. Richter, Timur Bagautdinov, Pierre Baque, and Pascal Fua. DeepMesh: Differentiable Iso- Surface Extraction. IEEE Trans. Pattern Anal. Mach. Intell., pages 1–15, 2024. 2

    work page 2024

  4. [4]

    Nondistorting flattening maps and the 3-D visualization of colon CT images

    S Halier, Sigurd Angenent, A Tannenbaurn, and Ron Kikinis. Nondistorting flattening maps and the 3-D visualization of colon CT images. IEEE Trans. Med. Imag., 19(7):665–670,

    work page

  5. [5]

    Diffeomorphic image registration with neural ve- locity field

    Kun Han, Shanlin Sun, Xiangyi Yan, Chenyu You, Hao Tang, Junayed Naushad, Haoyu Ma, Deying Kong, and Xi- aohui Xie. Diffeomorphic image registration with neural ve- locity field. In Proc. IEEE/CVF Winter Conf. Appl. Comp. Vis., pages 1869–1879, 2023. 3, 6

    work page 2023

  6. [6]

    MIPS: An efficient global parametrization method

    Kai Hormann and G ¨unther Greiner. MIPS: An efficient global parametrization method. Curve and Surface Design , pages 153–162, 2000. 2

    work page 2000

  7. [7]

    Texturenet: Consistent local parametrizations for learning from high- resolution signals on meshes

    Jingwei Huang, Haotian Zhang, Li Yi, Thomas Funkhouser, Matthias Nießner, and Leonidas J Guibas. Texturenet: Consistent local parametrizations for learning from high- resolution signals on meshes. In Proc. IEEE/CVF Conf. Comp. Vis. Pattern Recognit. (CVPR) , pages 4440–4449,

    work page

  8. [8]

    CPR-curved planar refor- mation

    Armin Kanitsar, Dominik Fleischmann, Rainer Wegenkittl, Petr Felkel, and Eduard Groller. CPR-curved planar refor- mation. IEEE, 2002. 2

    work page 2002

Show all 49 references
  1. [9]

    Advanced curved planar reformation: Flattening of vascular structures

    Armin Kanitsar, Rainer Wegenkittl, Dominik Fleischmann, and Eduard Groller. Advanced curved planar reformation: Flattening of vascular structures. IEEE, 2003. 2

    2003

  2. [10]

    Accelerating neural field training via soft mining

    Shakiba Kheradmand, Daniel Rebain, Gopal Sharma, Hossam Isack, Abhishek Kar, Andrea Tagliasacchi, and Kwang Moo Yi. Accelerating neural field training via soft mining. In Proc. IEEE/CVF Conf. Comp. Vis. Pattern Recog- nit. (CVPR), pages 20071–20080, 2024. 7

    2024

  3. [11]

    Clean implicit 3D structure from noisy 2D stem images

    Hannah Kniesel, Timo Ropinski, Tim Bergner, Kavitha Shaga Devan, Clarissa Read, Paul Walther, Tobias Ritschel, and Pedro Hermosilla. Clean implicit 3D structure from noisy 2D stem images. In Proc. IEEE/CVF Conf. Comp. Vis. Pattern Recognit. (CVPR), pages 20762–20772,

  4. [12]

    Kreiser, M

    J. Kreiser, M. Meuschke, G. Mistelbauer, B. Preim, and T. Ropinski. A Survey of Flattening-Based Medical Visualiza- tion Techniques. Computer Graphics Forum, 2018. 2

    2018

  5. [13]

    ADR- anatomy-driven reformation

    Jan Kretschmer, Grzegorz Soza, Christian Tietjen, Michael Suehling, Bernhard Preim, and Marc Stamminger. ADR- anatomy-driven reformation. IEEE Trans. Vis. Comp. Graphics, 20(12):2496–2505, 2014. 2, 5, 7, 8

    2014

  6. [14]

    A Local/Global Approach to Mesh Parameteriza- tion

    Ligang Liu, Lei Zhang, Yin Xu, Craig Gotsman, and Steven J Gortler. A Local/Global Approach to Mesh Parameteriza- tion. Computer Graphics Forum, 27:1495–1504, 2008. 2

    2008

  7. [15]

    DA W AND: Distortion-aware selection using neu- ral mesh parameterization

    Richard Liu, Noam Aigerman, Vladimir G Kim, and Rana Hanocka. DA W AND: Distortion-aware selection using neu- ral mesh parameterization. In Proc. IEEE/CVF Conf. Comp. Vis. Pattern Recognit. (CVPR), pages 16739–16749, 2023. 2

    2023

  8. [16]

    Least Squares Conformal Maps for Automatic Tex- ture Atlas Generation

    Bruno L ´evy, Sylvain Petitjean, Nicolas Ray, and J ´erome Maillot. Least Squares Conformal Maps for Automatic Tex- ture Atlas Generation. ACM Trans. Graph. , 21:362–371,

  9. [17]

    MeshFeat: Multi-Resolution Features for Neural Fields on Meshes

    Mihir Mahajan, Florian Hofherr, and Daniel Cremers. MeshFeat: Multi-Resolution Features for Neural Fields on Meshes. arXiv preprint arXiv:2407.13592, 2024. 2

    2024 arXiv

  10. [18]

    Bone Fracture and Lesion Assessment using Shape- Adaptive Unfolding

    Hannes Martinke, Christian Petry, Stefan Großkopf, Michael Suehling, Grzegorz Soza, Bernhard Preim, and Gabriel Mis- telbauer. Bone Fracture and Lesion Assessment using Shape- Adaptive Unfolding. VCBM, 17:149–158, 2017. 2

    2017

  11. [19]

    Implicit neural representation in medical imaging: A comparative survey

    Amirali Molaei, Amirhossein Aminimehr, Armin Tavakoli, Amirhossein Kazerouni, Bobby Azad, Reza Azad, and Dorit Merhof. Implicit neural representation in medical imaging: A comparative survey. In Proc. IEEE/CVF Int. Conf. Comp. Vis. (CVPR), pages 2381–2391, 2023. 3

    2023

  12. [20]

    Neural surface maps

    Luca Morreale, Noam Aigerman, Vladimir G Kim, and Niloy J Mitra. Neural surface maps. In Proc. IEEE/CVF Conf. Comp. Vis. Pattern Recognit. (CVPR) , pages 4639– 4648, 2021. 2

    2021

  13. [21]

    Spectral conformal parameterization

    Patrick Mullen, Yiying Tong, Pierre Alliez, and Mathieu Desbrun. Spectral conformal parameterization. In Computer Graphics Forum, pages 1487–1494, 2008. 2

    2008

  14. [22]

    Integrated visualization of morphologic and perfusion data for the analysis of coronary artery disease

    Steffen Oeltze, Anja Kuß, Frank Grothues, Anja Hennemuth, and Bernhard Preim. Integrated visualization of morphologic and perfusion data for the analysis of coronary artery disease. In EuroVis, pages 131–138, 2006. 2

    2006

  15. [23]

    Nerfmeshing: Distilling neural radiance fields into geometrically-accurate 3D meshes

    Marie-Julie Rakotosaona, Fabian Manhardt, Diego Martin Arroyo, Michael Niemeyer, Abhijit Kundu, and Federico Tombari. Nerfmeshing: Distilling neural radiance fields into geometrically-accurate 3D meshes. In Int. Conf. 3D Vis. (3DV), pages 1156–1165. IEEE, 2024. 2

    2024

  16. [24]

    9 The skull unfolded: a cranial CT visualization algorithm for fast and easy detection of skull fractures

    Helmut Ringl, Ruediger E Schernthaner, Gerd Schueller, Csilla Balassy, Daniela Kienzl, Ana Botosaneanu, Michael Weber, Christian Czerny, Stefan Hajdu, Thomas Mang, et al. 9 The skull unfolded: a cranial CT visualization algorithm for fast and easy detection of skull fractures....

    2010

  17. [25]

    Bifurcation matching for consistent cerebral vessel labeling in CTA of stroke patients

    Leonhard Rist, Oliver Taubmann, Florian Thamm, Hendrik Ditt, Michael S ¨uhling, and Andreas Maier. Bifurcation matching for consistent cerebral vessel labeling in CTA of stroke patients. Int. J. Comp. Assisted Radiol. Surgery , 18: 509 – 516, 2022. 2

    2022

  18. [26]

    Flexible Unfolding of Circu- lar Structures for Rendering Textbook-Style Cerebrovascular Maps

    Leonhard Rist, Oliver Taubmann, Hendrik Ditt, Michael S¨uhling, and Andreas Maier. Flexible Unfolding of Circu- lar Structures for Rendering Textbook-Style Cerebrovascular Maps. In Med. Image Comp. Comp. Assisted Intervention (MICCAI), pages 737–746, 2023. 2, 4, 5, 6, 8

    2023

  19. [27]

    Pancreatic vessel landmark detection in CT Angiography using prior anatomical knowledge

    Leonhard Rist, Christopher Homm, Felix Lades, Abra- ham Ayala Hernandez, Michael S¨uhling, Erik Gudman Steu- ble Brandt, Andreas Maier, and Oliver Taubmann. Pancreatic vessel landmark detection in CT Angiography using prior anatomical knowledge. In Artif. Intell. Pancreatic Di...

    2025

  20. [28]

    Texture mapping progressive meshes

    Pedro V Sander, John Snyder, Steven J Gortler, and Hugues Hoppe. Texture mapping progressive meshes. In Proc. 28th Annu. Conf. Comp. Graphics Interactive Techn., pages 409– 416, 2001. 2

    2001

  21. [29]

    Mesh param- eterization methods and their applications

    Alla Sheffer, Emil Praun, Kenneth Rose, et al. Mesh param- eterization methods and their applications. Found. Trends Comp. Graphics Vis., 2(2):105–171, 2007. 2

    2007

  22. [30]

    Martel, Alexander W

    Vincent Sitzmann, Julien N.P. Martel, Alexander W. Bergman, David B. Lindell, and Gordon Wetzstein. Implicit Neural Representations with Periodic Activation Functions. In Proc. NeurIPS, 2020. 3

    2020

  23. [31]

    Bijective parameterization with free boundaries

    Jason Smith and Scott Schaefer. Bijective parameterization with free boundaries. ACM Trans. Graphics (TOG), 34(4): 1–9, 2015. 2

    2015

  24. [32]

    As-rigid-as-possible surface modeling

    Olga Sorkine and Marc Alexa. As-rigid-as-possible surface modeling. In Symp. Geometry Process. , pages 109–116,

  25. [33]

    NISF: Neural implicit segmen- tation functions

    Nil Stolt-Ans ´o, Julian McGinnis, Jiazhen Pan, Kerstin Ham- mernik, and Daniel Rueckert. NISF: Neural implicit segmen- tation functions. In Int. Conf. Med. Image Comput. Comp.- Assisted Intervention (MICCAI), pages 734–744, 2023. 3

    2023

  26. [34]

    Duncan, and Xiaohui Xie

    Shanlin Sun, Kun Han, Chenyu You, Hao Tang, Deying Kong, Junayed Naushad, Xiangyi Yan, Haoyu Ma, Pooya Khosravi, James S. Duncan, and Xiaohui Xie. Medical im- age registration via neural fields. Med. Image Anal. , 97: 103249, 2024. 3, 4, 6, 7, 8

    2024

  27. [35]

    CoV- iCAD: Comprehensive visualization of coronary artery dis- ease

    Maurice Termeer, Javier Oliv ´an Besc ´os, Marcel Breeuwer, Anna Vilanova, Frans Gerritsen, and Eduard Gr ¨oller. CoV- iCAD: Comprehensive visualization of coronary artery dis- ease. IEEE Trans. Vis. Comp. Graphics , 13(6):1632–1639,

  28. [36]

    Optimization method for creating semi-isometric flat maps of the cerebral cortex

    Bijan Timsari and Richard M Leahy. Optimization method for creating semi-isometric flat maps of the cerebral cortex. In Med. Imag. 2000: Image Process., pages 698–708. SPIE,

    2000

  29. [37]

    Implicit neural representations for deformable image registration

    Jelmer M Wolterink, Jesse C Zwienenberg, and Christoph Brune. Implicit neural representations for deformable image registration. In Int. Conf. Med. Imag. Deep Learn. , pages 1349–1359. PMLR, 2022. 3, 4, 6

    2022

  30. [38]

    An arbitrary scale super- resolution approach for 3D MR images via implicit neu- ral representation

    Qing Wu, Yuwei Li, Yawen Sun, Yan Zhou, Hongjiang Wei, Jingyi Yu, and Yuyao Zhang. An arbitrary scale super- resolution approach for 3D MR images via implicit neu- ral representation. IEEE J. Biomed. Health Inform. , 27(2): 1004–1015, 2022. 3

    2022

  31. [39]

    NODEO: A Neural Or- dinary Differential Equation Based Optimization Framework for Deformable Image Registration

    Yifan Wu, Tom Z Jiahao, Jiancong Wang, Paul A Yushke- vich, M Ani Hsieh, and James C Gee. NODEO: A Neural Or- dinary Differential Equation Based Optimization Framework for Deformable Image Registration. In Proc. IEEE/CVF Conf. Comp. Vis. Pattern Recognit. (CVPR) , pages 20804–...

    2022

  32. [40]

    Ribseg dataset and strong point cloud baselines for rib segmentation from ct scans

    Jiancheng Yang, Shixuan Gu, Donglai Wei, Hanspeter Pfis- ter, and Bingbing Ni. Ribseg dataset and strong point cloud baselines for rib segmentation from ct scans. In Med. Image Comp. Comp. Assisted Intervention (MICCAI) , pages 611– 621, 2021. 2

    2021

  33. [41]

    Qi Zhang, Roy Eagleson, and Terry M. Peters. V olume vi- sualization: A technical overview with a focus on medical applications. J. Digit. Imag., 24:640–664, 2010. 1

    2010

  34. [42]

    Flattening-Net: Deep Regular 2D Rep- resentation for 3D Point Cloud Analysis

    Qijian Zhang, Junhui Hou, Yue Qian, Yiming Zeng, Juyong Zhang, and Ying He. Flattening-Net: Deep Regular 2D Rep- resentation for 3D Point Cloud Analysis. IEEE Trans. Pat- tern Anal. Mach. Intell., 45(8):9726–9742, 2023. 2

    2023

  35. [43]

    Flatten Anything: Unsupervised Neural Surface Parameteri- zation

    Qijian Zhang, Junhui Hou, Wenping Wang, and Ying He. Flatten Anything: Unsupervised Neural Surface Parameteri- zation. arXiv preprint arXiv:2405.14633, 2024. 3

    2024 arXiv

  36. [44]

    A review of 3D point clouds parameterization meth- ods

    Zaiping Zhu, Andres Iglesias, Lihua You, and Jian Jun Zhang. A review of 3D point clouds parameterization meth- ods. In Int. Conf. Comput. Sci., pages 690–703, 2022. 2 10 Neural Image Unfolding: Flattening Sparse Anatomical Structures using Neural Fields Supplementary Material

    2022

  37. [45]

    if GX g ||f e2 (ˆ xg) − f e1 (ˆ xg)|| < ϵ

    Stopping Criterion To assess early convergence, instead of monitoring the de- crease of the mutually influencing losses, we propose to stop fitting when the change in the output positions of G uniformly sampled inputs from the current epoch e2 is smaller than ϵ compared to an ...

  38. [46]

    Image Loss formulations Loss formulation for the pancreatic task with V being the binary pancreas organ mask: Lim = 1 S SX s 1 − V (ˆ xs) (7) Weighting function for vessel intensity image-loss with vmean as mean vessel intensity from the target points, k = 0.06 and l = 200: h(...

  39. [47]

    Distortion examples when the distance between neigh- boring read-out points does not match the image spacing

    Image Distortion Example (a) No distortion: Uniformly sampled read-out Volume Space Image Space (b) Stretching: Dense read-out in image center (c) Contraction: Sparse read-out in image center read-out positions Figure 10. Distortion examples when the distance between neigh- bo...

  40. [48]

    Data specifications for unfolded structures

    Data Table 1. Data specifications for unfolded structures. The patient data was either publicly available or collected in retrospective studies which received Institutional Review Board approval. The respective need for informed consent was either given or waived. Rib cage dat...

  41. [49]

    Additional Quantitative Results ARAP Ours CeVasMap Cerebral vessels 2 0 -2 -1 Pixelwise distortion in mm 2 0 -2 -1 Pixelwise distortion in mm Pancreas + vessels Hand bones Rib cage 0 . 2 5 2 . 7 5 . 9 0 . 2 3 9 . 5 1 4 . 1 0 . 3 4 4 . 2 1 5 . 2 0 . 3 4 6 . 5 1 3 . 5 0 . 2 4 5 ...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.