Pith. sign in

REVIEW 4 major objections 5 minor 82 references

CABLD: Contrast-Agnostic Brain Landmark Detection with Consistency-Based Regularization

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 self-supervised framework detects 32 brain landmarks in unseen MRI contrasts using only a single annotated template.

desk verdict A practical self-supervised landmark detector with a genuinely nice contrast-augmentation trick, but the abstract over-claims on T2w and the supervised baseline is compromised. read the letter →

arxiv 2411.17845 v3 pith:KGMH6OXS submitted 2024-11-26 eess.IV cs.CV

Soorena Salari , Arash Harirpoush , Hassan Rivaz , Yiming Xiao This is my paper · ORCID
classification eess.IVcs.CV
keywords brainlandmarkdetectionself-supervisedlearningmagneticresonanceimagingcontrast-agnosticrandomconvolutionconsistencyimageregistrationAFIDsprotocol
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

CABLD aims to detect standardized anatomical landmarks in 3D brain MRI without large labeled datasets. It trains a 3D CNN to predict the 32 AFIDs landmark positions using a single annotated template (ICBM152) plus unlabeled scans, by fitting thin-plate splines from each predicted landmark set to the template and checking whether the resulting warp aligns brain volumes. Two consistency losses, one across sampled subjects and one between subject and template, pull the predicted landmarks into a fixed protocol, while a 3D random-convolution augmentation teaches the model to ignore contrast differences. The paper reports mean radial errors around 3 to 5 mm across four datasets, including T2w scans never seen in training, and says the method outperforms classical registration, keypoint-based learning methods, and a fully supervised CNN.

What carries the argument

The load-bearing mechanism is a differentiable thin-plate spline: at every training step the network's current predicted landmarks are used to fit a TPS warp to the template landmarks (with a regularization parameter drawn from a log-uniform distribution), and the same warp feeds both the registration loss and the landmark consistency losses. A second component is 3D random convolution with 1x1x1 kernels and LeakyReLU activations, which creates many artificial contrast variants from each T1w scan so that the model learns contrast-invariant landmark features while keeping the monomodal MSE loss usable. An adaptive mixing coefficient alpha, growing from near zero to one over training, shifts emphasis from registration to consistency in a curriculum-learning style.

What would settle it

A reader could retrain CABLD on the same unlabeled T1w scans but replace the two consistency losses with direct regression to the 32 template landmarks on a small labeled subset; if direct supervision produces lower mean radial error, the paper's single-template sufficiency claim is weakened. Alternatively, re-running with the curriculum schedule inverted, putting consistency first and registration second, would test whether the reported convergence and accuracy actually depend on the adaptive ordering.

Watch

Extended reading notes

Core claim

The paper's central claim is that registration and landmark consistency reinforce each other well enough to bootstrap a clinically meaningful landmark detector from a single annotated reference. The registration loss gives the network anatomical context by warping each subject scan to the template and comparing voxel intensities, while the two consistency losses force the predicted landmark positions to agree in template space across subjects and with the template itself. With an adaptive schedule that starts with registration and then shifts to consistency, the model learns to place the 32 AFIDs landmarks in T1w scans and generalizes to T2w scans without any T2w labels. The authors argue that unlike keypoint methods driven purely by registration similarity, CABLD produces landmarks that follow a predefined anatomical protocol rather than landmarks that merely happen to align scans.

Load-bearing premise

The whole scheme rests on the assumption that a thin-plate-spline warp fitted from the model's own initially rough landmark predictions yields a registration loss that is a useful learning signal, so the self-supervised bootstrap converges before the consistency terms take over.

Editorial extensions

If this is right

  • Anatomical landmark detection in brain MRI can be done with one annotated template instead of hundreds of expert-labeled scans, lowering the barrier for new landmark protocols.
  • A model trained only on T1w data can localize landmarks in T2w scans when contrast augmentation is included, which matters for clinical sites with mixed MRI protocols.
  • Consistency-based regularization plus registration produces landmarks that follow a pre-defined protocol, unlike registration-driven keypoints that can drift across subjects.
  • The detected landmarks retain clinical utility: the paper's downstream experiments use inter-landmark distances to separate Parkinson's disease and Alzheimer's disease cohorts with reported F1 scores above 80 percent.
  • The method tolerates large added rotation misalignments with little degradation, suggesting it can handle the varying head positions common in real acquisitions.

Reading between the lines

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

  • Because the random-convolution augmentation is purely intensity-based, the same training recipe could plausibly transfer to other contrasts such as FLAIR or PD-weighted MRI without collecting paired multi-contrast data; this is a direct test a reader could run.
  • The implicit coordinate system created by the template-anchored consistency loss might make CABLD useful as a fast quality-control check for subject-to-template registration, since the landmarks are protocol-consistent and cheap to compute.
  • The curriculum schedule is itself an empirical claim: someone could vary the alpha schedule or remove the registration term after warmup and measure whether convergence and final MRE change, which would reveal how necessary the two-stage balance really is.
  • If pathological brains preserve the same gross anatomy, the method may also work on diseased scans without retraining, but that requires validation beyond the healthy subjects used for the main tables.
Share X Bluesky LinkedIn Reddit HN

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 CABLD, a self-supervised framework for 3D brain landmark detection that uses a single annotated template and unlabeled training scans. The method combines an inter-subject landmark consistency loss with an image registration loss, applies random-convolution contrast augmentation, and uses a curriculum-scheduled mixture of the two losses. Experiments are reported on four datasets (HCP T1w, OASIS, SNSX, and HCP T2w) using the AFIDs landmark protocol, with claims of state-of-the-art accuracy.

Significance. If the claims hold, the approach would substantially reduce annotation cost for landmark detection and offer a way to generalize to unseen MRI contrasts. The public code release, the use of the well-defined AFIDs protocol, and the breadth of evaluation datasets are strengths. However, the headline claims of universal state-of-the-art performance are only partially supported by the reported evidence, and several methodological details require clarification or correction.

major comments (4)
  1. [Section 4.4, Table 2, Abstract] The universal SOTA claim is contradicted by the T2w results. On HCP-T2w, CABLD has a higher (worse) MRE than ANTs(MI) (3.99±2.25 mm vs 3.91±2.19 mm) and a lower SDR at 3 mm (27.19% vs 35.00%; MultiGradICON also reaches 33.33%). CABLD is best only at SDR@6mm and SDR@9mm. Section 4.4 acknowledges the SDR@3mm deficit as 'slightly lower' but does not acknowledge the MRE deficit. Although Section 4.3 states that paired t-tests were used, no p-values are reported for these comparisons. Since HCP-T2w is the only evidence for contrast-agnostic generalization, the abstract's claim that CABLD 'outperforms the state-of-the-art methods in terms of MREs and SDRs' should be revised to a dataset-specific, significance-tested statement.
  2. [Section 4.3 and Section 4.1] The fully supervised baseline appears to be trained on the labeled test data. Section 4.1 defines the testing data as 122 scans from four sources and states that for each scan, 32 AFIDs landmarks were manually labeled. Section 4.3 says the supervised CNN is trained 'using the labeled datasets described in Sec. 4.1'. If the labeled test scans were used for training, the comparison with CABLD is invalid because the baseline had access to the test labels, making the claim that CABLD 'outperform[s] the supervised 3D CNN with statistical significance' misleading. Please clarify the exact data split and, if the supervised CNN was trained on test labels, retrain it with a disjoint train/test partition.
  3. [Section 3.2, Eqs. (1)-(5)] The consistency losses are self-referential and may be degenerate. In Eq. (1), β* is chosen to minimize Σ_j (T_β(f(x_i;θ))(j) - P(j))^2 + λI; Eq. (4) then evaluates exactly the first part of that minimized objective at β*, i.e., L_consistency2 is the squared residual of the regularized TPS fit against P. For λ=0, TPS interpolation makes this term exactly zero for any prediction; for small λ, it measures only the residual of a fit that was designed to minimize it. L_consistency1 (Eq. 3) inherits the same issue because both warps map their predictions near P. The paper does not report the actual distribution for sampling λ (a 'log-uniform distribution ranging between 0 and 10' is not well-defined), nor any analysis of the gradients. The mechanism claimed for 'inter-subject landmark consistency' is therefore not established. Please reformulate the consistency loss so it provides an independent constraint (e.g., by penalizing predicted-landmark distances to template-space locations without using a warp fitted to those same predictions) or provide a rigorous analysis of why the current formulation is non-degenerate in practice.
  4. [Section 3.4 and Section 4.2] The random convolution augmentation uses 1×1×1 kernels on single-channel input, which reduces to a per-voxel pointwise transformation (affine scaling plus bias followed by LeakyReLU) rather than a spatial convolution that mixes neighboring structures. The manuscript characterizes this as '3D convolution-based contrast augmentation' capable of modeling 'complex, non-linear intensity relationships'; as written, the augmentation is closer to a soft histogram transform. Because the contrast-agnostic claim rests partly on this augmentation, please clarify the operational effect of 1×1 random convolutions and temper the novelty statement accordingly.
minor comments (5)
  1. [Eq. (1) and Supplementary] Equation (1) writes the regularizer as λI, while the supplementary material defines the bending energy as I_T; please use consistent notation.
  2. [Section 4.2] The phrase 'randomly sampled from a log-uniform distribution ranging between 0 and 10' is not mathematically well-defined because a log-uniform distribution cannot include 0; please specify the exact sampling range and base.
  3. [Section 3.2] There are several typos, including 'anotmical' and 'paramter', which should be corrected.
  4. [Table 3] In the ablation table, the parenthetical values such as '(-49.99)' are not labeled; please clarify that they are absolute reductions in MRE (mm) relative to the base model.
  5. [Figure 3] The y-axis of Figure 3 should specify that MRE is in millimeters and state that lower values are better, to avoid ambiguity.

Circularity Check

1 steps flagged · score 6.0 of 10

CABLD's subject-template consistency loss is tautological: the TPS warp is fit by minimizing exactly the distance that L_consistency2 then reports, leaving only the λ bending penalty informative; external evaluation is independent.

  1. self definitional [Section 3.2, Eqs. (1), (3)-(5) and Supplementary Eq. (S6)]
    "G(P, f(xi;θ)) = arg min_β Σ_{j=1}^{L} (Tβ(f(xi;θ)^{(j)})−P^{(j)})^2 + λI, (1) ... L_consistency2 = (1/M) Σ_{k=1}^{M} ‖ T_{β*_{ik}}(f(x_{ik};θ)) − P ‖^2, (4) ... L_consistency = L_consistency1 + L_consistency2, (5)"

    Eq. (1) defines β* as the minimizer of the squared distance between TPS-warped predictions and the template landmarks P. Eq. (4) then evaluates exactly that same squared distance after substituting the minimizing β*. For λ→0 the TPS interpolates the predictions onto P, so L_consistency2 is ≈0 for any non-degenerate prediction; the only non-zero content comes from the λ bending-energy regularizer. Thus L_consistency2 is not an independent subject-template consistency penalty but the residual of a fit whose target is the same P used to construct it. L_consistency1 collapses for the same reason: both TPS warps map their predictions near P, so their difference is ≈0.

full rationale

The one load-bearing circular step is the consistency loss in Section 3.2. Because G is defined as the regularized minimizer of the same landmark-distance objective that L_consistency2 reports, that term cannot provide independent anatomical landmark supervision; it mainly enforces TPS smoothness. This is a genuine self-definitional reduction. However, the paper's headline empirical claim is not circular: MRE/SDR are evaluated against independent expert AFIDs labels on four datasets, and no load-bearing uniqueness or existence argument is imported from the authors' prior work. The 3D random-convolution augmentation is attributed to external work (Xu et al., ICLR 2021) and is not used tautologically. The T2w performance caveats (e.g., CABLD MRE 3.99 mm vs ANTs(MI) 3.91 mm and lower SDR@3mm in Table 2) are correctness concerns, not circularity. On balance, the derivation is partially circular in one loss term but the final evaluation stands outside the fitted loop, giving a score of 6 rather than 8-10.

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

No new physical entities are introduced; the method is algorithmic. The free parameters listed are the main hand-chosen hyperparameters that shape the training signal. The axioms include the mathematical foundation of TPS and several domain assumptions about template validity, MSE similarity, contrast augmentation, and preprocessing that the central claim depends on.

free parameters (6)
  • TPS regularization weight lambda = sampled log-uniformly in [0,10] during training
    Controls how strictly the TPS interpolates predicted landmarks to template landmarks; when lambda is large the warp is near-affine and the consistency loss is not identically zero. Chosen by hand, not fit to data.
  • Curriculum schedule slope (5 in Eq. 7) = 5
    Sets the rate at which alpha rises from 0 to 1; hand-chosen and not ablated.
  • Random convolution configuration = 5 layers, kernel 1x1x1, weights U(0,2) zero-centered, LeakyReLU slope 0.2
    Defines the contrast augmentation family; chosen by the authors, no tuning search reported.
  • Consistency group size M = 2
    Number of scans used for inter-subject consistency; set to 2 for memory reasons, larger M not evaluated.
  • Geometric augmentation ranges = rotations [-180,180], translations [-15,15], scale [0.8,1.2], shear [-0.1,0.1]
    Hand-chosen augmentation bounds for affine augmentation.
  • Training epochs and optimizer schedule = 2500 epochs, Adam lr 1e-4 to 1e-6 cosine
    Training hyperparameters; no sensitivity analysis is reported.
assumptions (6)
  • standard math Thin-plate spline has a closed-form differentiable solution for landmark-based warp (supplementary Eqs. S1-S6).
    The method relies on TPS being differentiable so that gradients flow through the computed warp to the landmark network.
  • domain assumption A single template with 32 AFIDs landmarks defines a consistent anatomical landmark protocol for all subjects.
    The whole training signal assumes the template landmarks are valid for every subject and that cross-subject consistency can be measured against them.
  • domain assumption MSE between the warped subject scan and the template is a valid registration similarity after random-convolution augmentation.
    The registration loss in Eq. 2 uses MSE; this assumes RC-transformed T1w intensities are comparable to the T1w template voxel-wise.
  • domain assumption 1x1 random convolutions with LeakyReLU can simulate intensity relationships across MRI contrasts, including T1-to-T2.
    Central to the contrast-agnostic generalization claim; only tested empirically on HCP-T2w.
  • domain assumption Preprocessing affine registration to ICBM152 space provides sufficient gross alignment for TPS-based correspondence.
    All training and test scans are affinely registered to the template space before landmark learning; without this, the single-template consistency assumption likely fails.
  • ad hoc to paper The residual of the regularized TPS fit in Eq. 1 is a meaningful training signal rather than a mathematical artifact.
    L_consistency2 is computed from the same TPS fitted to minimize it; the paper relies on lambda regularization to make this residual nonzero and informative.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CABLD: Contrast-Agnostic Brain Landmark Detection with Consistency-Based Regularization." pith.science (2026). https://pith.science/paper/KGMH6OXS

@misc{pith2026241117845,
  author       = {Pith},
  title        = {Pith review of: CABLD: Contrast-Agnostic Brain Landmark Detection with Consistency-Based Regularization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KGMH6OXS}},
  note         = {Machine review of arXiv:2411.17845}
}
read the original abstract

Anatomical landmark detection in medical images is essential for various clinical and research applications, including disease diagnosis and surgical planning. However, manual landmark annotation is time-consuming and requires significant expertise. Existing deep learning (DL) methods often require large amounts of well-annotated data, which are costly to acquire. In this paper, we introduce CABLD, a novel self-supervised DL framework for 3D brain landmark detection in unlabeled scans with varying contrasts by using only a single reference example. To achieve this, we employed an inter-subject landmark consistency loss with an image registration loss while introducing a 3D convolution-based contrast augmentation strategy to promote model generalization to new contrasts. Additionally, we utilize an adaptive mixed loss function to schedule the contributions of different sub-tasks for optimal outcomes. We demonstrate the proposed method with the intricate task of MRI-based 3D brain landmark detection. With comprehensive experiments on four diverse clinical and public datasets, including both T1w and T2w MRI scans at different MRI field strengths, we demonstrate that CABLD outperforms the state-of-the-art methods in terms of mean radial errors (MREs) and success detection rates (SDRs). Our framework provides a robust and accurate solution for anatomical landmark detection, reducing the need for extensively annotated datasets and generalizing well across different imaging contrasts. Our code is publicly available at https://github.com/HealthX-Lab/CABLD.

Figures

Figures reproduced from arXiv: 2411.17845 by the authors.

Figure 1
Figure 1. Overview of the proposed framework and the consistency-based learning approach. For simplicity, 3D scans are presented in a [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Qualitative comparison of anatomical landmark detection results using the proposed technique, illustrated on coronal, axial, [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Comparison of different methods’ MRE (mm) perfor [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

82 extracted references · 79 canonical work pages

  1. [1]

    Application of the anatomical fiducials framework to a clinical dataset of patients with parkinson’s disease

    Mohamad Abbass, Greydon Gilmore, Alaa Taha, Ryan Chevalier, Magdalena Jach, Terry M Peters, Ali R Khan, and Jonathan C Lau. Application of the anatomical fiducials framework to a clinical dataset of patients with parkinson’s disease. Brain Structure and Function, pages 1–13, 2022. 1, 7, 4

    2022

  2. [2]

    The Amsterdam Ultra-high field adult lifespan database (AHEAD): A freely available multimodal 7 Tesla submillimeter magnetic resonance imaging database

    Anneke Alkemade, Martijn J Mulder, Josephine M Groot, Bethany R Isaacs, Nikita van Berendonk, Nicky Lute, Scott JS Isherwood, Pierre-Louis Bazin, and Birte U Forstmann. The Amsterdam Ultra-high field adult lifespan database (AHEAD): A freely available multimodal 7 Tesla submillimeter magnetic resonance imaging database. Neu- roImage, 221:117200, 2020. 5

    2020

  3. [3]

    Advanced normalization tools (ANTS)

    Brian B Avants, Nick Tustison, Gang Song, et al. Advanced normalization tools (ANTS). Insight j, 2(365):1–35, 2009. 2, 6, 7

    work page 2009

  4. [4]

    Curriculum learning

    Yoshua Bengio, J ´erˆome Louradour, Ronan Collobert, and Ja- son Weston. Curriculum learning. In International Confer- ence on Machine Learning, pages 41–48, 2009. 3, 4

    work page 2009

  5. [5]

    Self-supervised discovery of anatomical shape landmarks

    Riddhish Bhalodia, Ladislav Kavan, and Ross T Whitaker. Self-supervised discovery of anatomical shape landmarks. In Medical Image Computing and Computer Assisted, pages 627–638. Springer, 2020. 1, 2, 3, 8

    work page 2020

  6. [6]

    Leveraging unsupervised image registration for discovery of landmark shape descriptor

    Riddhish Bhalodia, Shireen Elhabian, Ladislav Kavan, and Ross Whitaker. Leveraging unsupervised image registration for discovery of landmark shape descriptor. Medical Image Analysis, 73:102157, 2021. 1, 2, 3, 8

    work page 2021

  7. [7]

    SE (3)-equivariant and noise-invariant 3D rigid mo- tion tracking in brain MRI

    Benjamin Billot, Neel Dey, Daniel Moyer, Malte Hoffmann, Esra Abaci Turk, Borjan Gagoski, P Ellen Grant, and Polina Golland. SE (3)-equivariant and noise-invariant 3D rigid mo- tion tracking in brain MRI. IEEE Transactions on Medical Imaging, 2024. 6

    work page 2024

  8. [8]

    Bookstein

    Fred L. Bookstein. Principal warps: Thin-plate splines and the decomposition of deformations. IEEE Transactions on Pattern Analysis and Machine Intelligence , 11(6):567–585,

    work page

Show all 82 references
  1. [9]

    3fabrec: Fast few- shot face alignment by reconstruction

    Bjorn Browatzki and Christian Wallraven. 3fabrec: Fast few- shot face alignment by reconstruction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6110–6120, 2020. 2, 3

    2020

  2. [10]

    Self-supervised landmark learning with deformation reconstruction and cross-subject consistency objectives

    Chun-Hung Chao and Marc Niethammer. Self-supervised landmark learning with deformation reconstruction and cross-subject consistency objectives. In International Work- shop on PRedictive Intelligence In MEdicine , pages 70–82. Springer, 2023. 2

    2023

  3. [11]

    A survey on deep learning in medical image reg- istration: New technologies, uncertainty, evaluation metrics, and beyond

    Junyu Chen, Yihao Liu, Shuwen Wei, Zhangxing Bian, Shalini Subramanian, Aaron Carass, Jerry L Prince, and Yong Du. A survey on deep learning in medical image reg- istration: New technologies, uncertainty, evaluation metrics, and beyond. Medical Image Analysis, page 103385, 2024. 2

    2024

  4. [12]

    Fast and ac- curate craniomaxillofacial landmark detection via 3D faster R-CNN

    Xiaoyang Chen, Chunfeng Lian, Hannah H Deng, Tianshu Kuang, Hung-Ying Lin, Deqiang Xiao, Jaime Gateno, Ding- gang Shen, James J Xia, and Pew-Thian Yap. Fast and ac- curate craniomaxillofacial landmark detection via 3D faster R-CNN. IEEE Transactions on Medical Imaging , 40(12)...

    2021

  5. [13]

    MultiGradICON: A Foundation Model for Multimodal Medical Image Registration

    Bas ¸ar Demir, Lin Tian, Hastings Greer, Roland Kwitt, Franc ¸ois-Xavier Vialard, Ra´ul San Jos ´e Est ´epar, Sylvain Bouix, Richard Rushmore, Ebrahim Ebrahim, and Marc Niethammer. MultiGradICON: A Foundation Model for Multimodal Medical Image Registration. In International Wo...

    2024

  6. [14]

    Approximate thin plate spline mappings

    Gianluca Donato and Serge Belongie. Approximate thin plate spline mappings. In Computer Vision—ECCV 2002: 7th European Conference on Computer Vision Copenhagen, Denmark, May 28–31, 2002 Proceedings, Part III 7 , pages 21–31. Springer, 2002. 1

    2002

  7. [15]

    Beast: brain extraction based on nonlocal segmentation technique

    Simon F Eskildsen, Pierrick Coup ´e, Vladimir Fonov, Jos´e V Manj´on, Kelvin K Leung, Nicolas Guizard, Shafik N Wassef, Lasse Riis Østergaard, D Louis Collins, Alzheimer’s Dis- ease Neuroimaging Initiative, et al. Beast: brain extraction based on nonlocal segmentation techniqu...

    2012

  8. [16]

    KeyMorph: Robust multi-modal affine registration via unsupervised keypoint detection

    M Yu Evan, Alan Q Wang, Adrian V Dalca, and Mert R Sabuncu. KeyMorph: Robust multi-modal affine registration via unsupervised keypoint detection. In Medical Imaging with Deep Learning, 2022. 1, 2, 8

    2022

  9. [17]

    Understanding vari- ability in brain MRI templates: Optimal sample sizes for rep- resentative population averages

    Vladimir S Fonov and D Louis Collins. Understanding vari- ability in brain MRI templates: Optimal sample sizes for rep- resentative population averages. Aperture Neuro, 5, 2025. 3, 4

    2025

  10. [18]

    Unbiased nonlinear aver- age age-appropriate brain templates from birth to adulthood

    Vladimir S Fonov, Alan C Evans, Robert C McKinstry, C Robert Almli, and DL Collins. Unbiased nonlinear aver- age age-appropriate brain templates from birth to adulthood. NeuroImage, 47:S102, 2009. 5

    2009

  11. [19]

    Symmetric at- lasing and model based segmentation: an application to the hippocampus in older adults

    G ¨unther Grabner, Andrew L Janke, Marc M Budge, David Smith, Jens Pruessner, and D Louis Collins. Symmetric at- lasing and model based segmentation: an application to the hippocampus in older adults. In International Conference on Medical Image Computing and Computer Assisted...

    2006

  12. [20]

    Synthmorph: learning contrast-invariant registration without acquired images

    Malte Hoffmann, Benjamin Billot, Douglas N Greve, Juan Eugenio Iglesias, Bruce Fischl, and Adrian V Dalca. Synthmorph: learning contrast-invariant registration without acquired images. IEEE Transactions on Medical Imaging , 41(3):543–558, 2021. 2

    2021

  13. [21]

    Enhancement of MR images using registration for signal averaging

    Colin J Holmes, Rick Hoge, Louis Collins, Roger Woods, Arthur W Toga, and Alan C Evans. Enhancement of MR images using registration for signal averaging. Journal of Computer Assisted Tomography, 22(2):324–333, 1998. 3

    1998

  14. [22]

    Improving landmark localization with semi-supervised learning

    Sina Honari, Pavlo Molchanov, Stephen Tyree, Pascal Vin- cent, Christopher Pal, and Jan Kautz. Improving landmark localization with semi-supervised learning. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 1546–1555, 2018. 3

    2018

  15. [23]

    Whitwell, Chadwick Ward, et al

    Clifford R Jack Jr, Matt A Bernstein, Nick C Fox, Paul Thompson, Gene Alexander, Danielle Harvey, Bret Borowski, Paula J Britson, Jennifer L. Whitwell, Chadwick Ward, et al. The Alzheimer’s disease neuroimaging initia- tive (ADNI): MRI methods. Journal of Magnetic Resonance Im...

    2008

  16. [24]

    Multi-task lo- calization and segmentation for X-Ray guided planning in knee surgery

    Florian Kordon, Peter Fischer, Maxim Privalov, Benedict Swartman, Marc Schnetzke, Jochen Franke, Ruxandra La- sowski, Andreas Maier, and Holger Kunze. Multi-task lo- calization and segmentation for X-Ray guided planning in knee surgery. In International Conference on Medical I...

    2019

  17. [25]

    Automatic detection of craniomaxillofacial anatomical landmarks on CBCT im- ages using 3D mask R-CNN

    Yankun Lang, Li Wang, Pew-Thian Yap, Chunfeng Lian, Hannah Deng, Kim-Han Thung, Deqiang Xiao, Peng Yuan, Steve GF Shen, Jaime Gateno, et al. Automatic detection of craniomaxillofacial anatomical landmarks on CBCT im- ages using 3D mask R-CNN. In International Workshop on Graph...

    2019

  18. [26]

    Automatic localization of landmarks in craniomaxillofacial CBCT images using a local attention-based graph convolution network

    Yankun Lang, Chunfeng Lian, Deqiang Xiao, Hannah Deng, Peng Yuan, Jaime Gateno, Steve GF Shen, David M Alfi, Pew-Thian Yap, James J Xia, et al. Automatic localization of landmarks in craniomaxillofacial CBCT images using a local attention-based graph convolution network. In In...

    2020

  19. [27]

    End-to-end coordinate regression model with attention- guided mechanism for landmark localization in 3D medical images

    Jupeng Li, Yinghui Wang, Junbo Mao, Gang Li, and Ruohan Ma. End-to-end coordinate regression model with attention- guided mechanism for landmark localization in 3D medical images. In International Workshop on Machine Learning in Medical Imaging, pages 624–633. Springer, 2020. 2

    2020

  20. [28]

    Multi-task dynamic transformer network for concurrent bone segmentation and large-scale landmark localization with dental CBCT

    Chunfeng Lian, Fan Wang, Hannah H Deng, Li Wang, De- qiang Xiao, Tianshu Kuang, Hung-Ying Lin, Jaime Gateno, Steve GF Shen, Pew-Thian Yap, et al. Multi-task dynamic transformer network for concurrent bone segmentation and large-scale landmark localization with dental CBCT. In ...

    2020

  21. [29]

    Landmark-based deep multi-instance learning for brain dis- ease diagnosis

    Mingxia Liu, Jun Zhang, Ehsan Adeli, and Dinggang Shen. Landmark-based deep multi-instance learning for brain dis- ease diagnosis. Medical Image Analysis, 43:157–168, 2018. 1, 7

    2018

  22. [30]

    Anatomical landmark based deep feature representation for MR images in brain disease diagnosis

    Mingxia Liu, Jun Zhang, Dong Nie, Pew-Thian Yap, and Dinggang Shen. Anatomical landmark based deep feature representation for MR images in brain disease diagnosis. IEEE Journal of Biomedical and Health Informatics , 22(5): 1476–1485, 2018

    2018

  23. [31]

    Anatomical-landmark-based deep learning for Alzheimer’s disease diagnosis with structural magnetic resonance imag- ing

    Mingxia Liu, Chunfeng Lian, and Dinggang Shen. Anatomical-landmark-based deep learning for Alzheimer’s disease diagnosis with structural magnetic resonance imag- ing. Deep Learning in Healthcare: Paradigms and Applica- tions, pages 127–147, 2020. 1, 7

    2020

  24. [32]

    V olumetric landmark detection with a multi-scale shift equivariant neu- ral network

    Tianyu Ma, Ajay Gupta, and Mert R Sabuncu. V olumetric landmark detection with a multi-scale shift equivariant neu- ral network. In 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI), pages 981–985. IEEE, 2020. 5

    2020

  25. [33]

    Open access series of imaging studies (OASIS): cross-sectional MRI data in young, middle aged, nondemented, and demented older adults

    Daniel S Marcus, Tracy H Wang, Jamie Parker, John G Cser- nansky, John C Morris, and Randy L Buckner. Open access series of imaging studies (OASIS): cross-sectional MRI data in young, middle aged, nondemented, and demented older adults. Journal of Cognitive Neuroscience , 19(9...

    2007

  26. [34]

    The Parkin- son progression marker initiative (PPMI)

    Kenneth Marek, Danna Jennings, Shirley Lasch, Andrew Siderowf, Caroline Tanner, Tanya Simuni, Chris Coffey, Karl Kieburtz, Emily Flagg, Sohini Chowdhury, et al. The Parkin- son progression marker initiative (PPMI). Progress in Neu- robiology, 95(4):629–635, 2011. 7

    2011

  27. [35]

    Contour-hugging heatmaps for landmark detection

    James McCouat and Irina V oiculescu. Contour-hugging heatmaps for landmark detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 20597–20605, 2022. 1, 2

    2022

  28. [36]

    Fast free-form deformation using graphics processing units

    Marc Modat, Gerard R Ridgway, Zeike A Taylor, Manja Lehmann, Josephine Barnes, David J Hawkes, Nick C Fox, and S ´ebastien Ourselin. Fast free-form deformation using graphics processing units. Computer Methods and Programs in Biomedicine, 98(3):278–284, 2010. 2, 6, 7

    2010

  29. [37]

    Modality-Agnostic Structural Image Representation Learn- ing for Deformable Multi-Modality Medical Image Registra- tion

    Tony CW Mok, Zi Li, Yunhao Bai, Jianpeng Zhang, Wei Liu, Yan-Jie Zhou, Ke Yan, Dakai Jin, Yu Shi, Xiaoli Yin, et al. Modality-Agnostic Structural Image Representation Learn- ing for Deformable Multi-Modality Medical Image Registra- tion. In Proceedings of the IEEE/CVF Conferen...

  30. [38]

    Unsupervised domain adaptation of MRI skull-stripping trained on adult data to newborns

    Abbas Omidi, Aida Mohammadshahi, Neha Gianchandani, Regan King, Lara Leijser, and Roberto Souza. Unsupervised domain adaptation of MRI skull-stripping trained on adult data to newborns. In Proceedings of the IEEE/CVF Win- ter Conference on Applications of Computer Vision , pag...

    2024

  31. [39]

    Unsuper- vised domain adaptation of brain MRI skull stripping trained on adult data to newborns: Combining synthetic data with domain invariant features

    Abbas Omidi, Amirmohammad Shamaei, Anouk Verschuu, Regan King, Lara Leijser, and Roberto Souza. Unsuper- vised domain adaptation of brain MRI skull stripping trained on adult data to newborns: Combining synthetic data with domain invariant features. In Medical Imaging with Dee...

    2024

  32. [40]

    Causality-inspired single- source domain generalization for medical image segmenta- tion

    Cheng Ouyang, Chen Chen, Surui Li, Zeju Li, Chen Qin, Wenjia Bai, and Daniel Rueckert. Causality-inspired single- source domain generalization for medical image segmenta- tion. IEEE Transactions on Medical Imaging , 42(4):1095– 1106, 2022. 2, 5

    2022

  33. [41]

    Which images to label for few-shot medical landmark detec- tion? In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 20606–20616,

    Quan Quan, Qingsong Yao, Jun Li, and S Kevin Zhou. Which images to label for few-shot medical landmark detec- tion? In Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition , pages 20606–20616,

  34. [42]

    Which images to label for few-shot medical image analysis? Medical Image Analysis, 96:103200, 2024

    Quan Quan, Qingsong Yao, Heqin Zhu, Qiyuan Wang, and S Kevin Zhou. Which images to label for few-shot medical image analysis? Medical Image Analysis, 96:103200, 2024. 1

    2024

  35. [43]

    V olumetric image registration from invariant keypoints

    Blaine Rister, Mark A Horowitz, and Daniel L Rubin. V olumetric image registration from invariant keypoints. IEEE Transactions on Image Processing, 26(10):4900–4910,

  36. [44]

    Image similarity and tissue overlaps as surrogates for image registration accuracy: widely used but unreliable

    Torsten Rohlfing. Image similarity and tissue overlaps as surrogates for image registration accuracy: widely used but unreliable. IEEE Transactions on Medical Imaging , 31(2): 153–163, 2011. 1

    2011

  37. [45]

    Landmark-based elas- tic registration using approximating thin-plate splines

    Karl Rohr, H Siegfried Stiehl, Rainer Sprengel, Thorsten M Buzug, J¨urgen Weese, and MH Kuhn. Landmark-based elas- tic registration using approximating thin-plate splines. IEEE Transactions on Medical Imaging, 20(6):526–534, 2001. 1

    2001

  38. [46]

    Uncertainty-aware transformer model for anatomical land- mark detection in paraspinal muscle MRIs

    Soorena Salari, Amirhossein Rasoulian, Michele Bat- tie, Maryse Fortin, Hassan Rivaz, and Yiming Xiao. Uncertainty-aware transformer model for anatomical land- mark detection in paraspinal muscle MRIs. InMedical Imag- ing 2023: Image Processing, pages 246–252. SPIE, 2023. 1

    2023

  39. [47]

    Age differences in the functional architecture of the human brain

    Roni Setton, Laetitia Mwilambwe-Tshilobo, Manesh Girn, Amber W Lockrow, Giulia Baracchini, Colleen Hughes, Alexander J Lowe, Benjamin N Cassidy, Jian Li, Wen-Ming Luh, et al. Age differences in the functional architecture of the human brain. Cerebral Cortex, 33(1):114–134, 2023. 5

    2023

  40. [48]

    Mitigating aberration-induced noise: A deep learning-based aberration-to-aberration ap- proach

    Mostafa Sharifzadeh, Sobhan Goudarzi, An Tang, Habib Benali, and Hassan Rivaz. Mitigating aberration-induced noise: A deep learning-based aberration-to-aberration ap- proach. IEEE Transactions on Medical Imaging, 2024. 4

    2024

  41. [49]

    MAD: Modality agnostic distance measure for image registration

    Vasiliki Sideri-Lampretsa, Veronika A Zimmer, Huaqi Qiu, Georgios Kaissis, and Daniel Rueckert. MAD: Modality agnostic distance measure for image registration. In In- ternational Conference on Medical Image Computing and Computer-Assisted Intervention , pages 147–156. Springer,

  42. [50]

    Data consistent deep rigid MRI motion cor- rection

    Nalini M Singh, Neel Dey, Malte Hoffmann, Bruce Fis- chl, Elfar Adalsteinsson, Robert Frost, Adrian V Dalca, and Polina Golland. Data consistent deep rigid MRI motion cor- rection. In Medical Imaging with Deep Learning, pages 368–

  43. [51]

    Fully convolutional regression network for accurate detection of measurement points

    Michal Sofka, Fausto Milletari, Jimmy Jia, and Alex Roth- berg. Fully convolutional regression network for accurate detection of measurement points. In Deep Learning in Med- ical Image Analysis and Multimodal Learning for Clinical Decision Support, pages 258–266. Springer, 2017. 5

    2017

  44. [52]

    Neurocognitive aging data release with behav- ioral, structural and multi-echo functional MRI measures

    R Nathan Spreng, Roni Setton, Udi Alter, Benjamin N Cas- sidy, Bri Darboh, Elizabeth DuPre, Karin Kantarovich, Am- ber W Lockrow, Laetitia Mwilambwe-Tshilobo, Wen-Ming Luh, et al. Neurocognitive aging data release with behav- ioral, structural and multi-echo functional MRI mea...

    2022

  45. [53]

    Magnetic resonance imag- ing datasets with anatomical fiducials for quality control and registration

    Alaa Taha, Greydon Gilmore, Mohamad Abbass, Jason Kai, Tristan Kuehn, John Demarco, Geetika Gupta, Chris Zajner, Daniel Cao, Ryan Chevalier, et al. Magnetic resonance imag- ing datasets with anatomical fiducials for quality control and registration. Scientific Data, 10(1):449,...

    2023

  46. [54]

    uniGradICON: A Foun- dation Model for Medical Image Registration

    Lin Tian, Hastings Greer, Roland Kwitt, Franc ¸ois-Xavier Vialard, Ra ´ul San Jos ´e Est ´epar, Sylvain Bouix, Richard Rushmore, and Marc Niethammer. uniGradICON: A Foun- dation Model for Medical Image Registration. In In- ternational Conference on Medical Image Computing and ...

  47. [55]

    Joint training of a convolutional network and a graphical model for human pose estimation

    Jonathan J Tompson, Arjun Jain, Yann LeCun, and Christoph Bregler. Joint training of a convolutional network and a graphical model for human pose estimation. Advances in Neural Information Processing Systems, 27, 2014. 1

    2014

  48. [56]

    Deep geodesic learn- ing for segmentation and anatomical landmarking

    Neslisah Torosdagli, Denise K Liberton, Payal Verma, Murat Sincan, Janice S Lee, and Ulas Bagci. Deep geodesic learn- ing for segmentation and anatomical landmarking. IEEE Transactions on Medical Imaging, 38(4):919–931, 2018. 3

    2018

  49. [57]

    Automatic point landmark matching for regulariz- ing nonlinear intensity registration: Application to thoracic CT images

    Martin Urschler, Christopher Zach, Hendrik Ditt, and Horst Bischof. Automatic point landmark matching for regulariz- ing nonlinear intensity registration: Application to thoracic CT images. In International Conference on Medical Image Computing and Computer-Assisted Interventi...

  50. [58]

    The WU-Minn human con- nectome project: an overview

    David C Van Essen, Stephen M Smith, Deanna M Barch, Timothy EJ Behrens, Essa Yacoub, Kamil Ugurbil, Wu- Minn HCP Consortium, et al. The WU-Minn human con- nectome project: an overview. Neuroimage, 80:62–79, 2013. 5

    2013

  51. [59]

    A robust and interpretable deep learning frame- work for multi-modal registration via keypoints

    Alan Q Wang, M Yu Evan, Adrian V Dalca, and Mert R Sabuncu. A robust and interpretable deep learning frame- work for multi-modal registration via keypoints. Medical Image Analysis, 90:102962, 2023. 1, 2, 3, 6, 7, 8

    2023

  52. [60]

    BrainMorph: A Foun- dational Keypoint Model for Robust and Flexible Brain MRI Registration

    Alan Q Wang, Rachit Saluja, Heejong Kim, Xinzi He, Adrian Dalca, and Mert R Sabuncu. BrainMorph: A Foun- dational Keypoint Model for Robust and Flexible Brain MRI Registration. Machine Learning for Biomedical Imaging, 3: 181–203, 2025. 1, 2, 3, 6, 7, 8

    2025

  53. [61]

    A survey on curriculum learning

    Xin Wang, Yudong Chen, and Wenwu Zhu. A survey on curriculum learning. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(9):4555–4576, 2021. 3, 4

    2021

  54. [62]

    CAR: Contrast-agnostic deformable medical im- age registration with contrast-invariant latent regularization

    Yinsong Wang, Siyi Du, Shaoming Zheng, Xinzhe Luo, and Chen Qin. CAR: Contrast-agnostic deformable medical im- age registration with contrast-invariant latent regularization. In International Workshop on Biomedical Image Registra- tion, pages 308–318. Springer, 2024. 2, 5

    2024

  55. [63]

    Image guidance in deep brain stimulation surgery to treat Parkin- son’s disease: A comprehensive review

    Yiming Xiao, Jonathan C Lau, Dimuthu Hemachandra, Greydon Gilmore, Ali R Khan, and Terry M Peters. Image guidance in deep brain stimulation surgery to treat Parkin- son’s disease: A comprehensive review. IEEE Transactions on Biomedical Engineering, 68(3):1024–1033, 2020. 1

    2020

  56. [64]

    A population-averaged struc- tural connectomic brain atlas dataset from 422 HCP-aging subjects

    Yiming Xiao, Greydon Gilmore, Jason Kai, Jonathan C Lau, Terry Peters, and Ali R Khan. A population-averaged struc- tural connectomic brain atlas dataset from 422 HCP-aging subjects. Data in Brief, 50:109513, 2023. 2

    2023

  57. [65]

    Robust and generalizable visual representation learning via random convolutions

    Zhenlin Xu, Deyi Liu, Junlin Yang, Colin Raffel, and Marc Niethammer. Robust and generalizable visual representation learning via random convolutions. In International Confer- ence on Learning Representations, 2021. 5

    2021

  58. [66]

    Vertebra-focused landmark detection for scoliosis assessment

    Jingru Yi, Pengxiang Wu, Qiaoying Huang, Hui Qu, and Dimitris N Metaxas. Vertebra-focused landmark detection for scoliosis assessment. In 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI) , pages 736–740. IEEE, 2020. 1

    2020

  59. [67]

    Gradient surgery for multi-task learning

    Tianhe Yu, Saurabh Kumar, Abhishek Gupta, Sergey Levine, Karol Hausman, and Chelsea Finn. Gradient surgery for multi-task learning. Advances in Neural Information Pro- cessing Systems, 33:5824–5836, 2020. 4

    2020

  60. [68]

    Learning-based local-to-global land- mark annotation for automatic 3D cephalometry

    Hye Sun Yun, Tae Jun Jang, Sung Min Lee, Sang-Hwy Lee, and Jin Keun Seo. Learning-based local-to-global land- mark annotation for automatic 3D cephalometry. Physics in Medicine & Biology, 65(8):085018, 2020. 2

    2020

  61. [69]

    Alzheimer’s disease diagnosis using landmark- based features from longitudinal structural MR images

    Jun Zhang, Mingxia Liu, Le An, Yaozong Gao, and Ding- gang Shen. Alzheimer’s disease diagnosis using landmark- based features from longitudinal structural MR images. IEEE Journal of Biomedical and Health Informatics , 21(6): 1607–1616, 2017. 1, 7

    2017

  62. [70]

    Landmark-based Alzheimer’s disease diagno- sis using longitudinal structural MR images

    Jun Zhang, Mingxia Liu, Le An, Yaozong Gao, and Ding- gang Shen. Landmark-based Alzheimer’s disease diagno- sis using longitudinal structural MR images. In Medical Computer Vision and Bayesian and Graphical Models for Biomedical Imaging, pages 35–45. Springer, 2017. 1, 7

    2017

  63. [71]

    Detect- ing anatomical landmarks from limited medical imaging data using two-stage task-oriented deep neural networks

    Jun Zhang, Mingxia Liu, and Dinggang Shen. Detect- ing anatomical landmarks from limited medical imaging data using two-stage task-oriented deep neural networks. IEEE Transactions on Image Processing, 26(10):4753–4764,

  64. [72]

    Context-guided fully convolu- tional networks for joint craniomaxillofacial bone segmen- tation and landmark digitization

    Jun Zhang, Mingxia Liu, Li Wang, Si Chen, Peng Yuan, Jianfu Li, Steve Guo-Fang Shen, Zhen Tang, Ken-Chung Chen, James J Xia, et al. Context-guided fully convolu- tional networks for joint craniomaxillofacial bone segmen- tation and landmark digitization. Medical Image Analysis...

    2020

  65. [73]

    Thin-plate spline motion model for image animation

    Jian Zhao and Hui Zhang. Thin-plate spline motion model for image animation. In Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, pages 3657–3666, 2022. 1

    2022

  66. [74]

    UOD: Universal one-shot detection of anatomical landmarks

    Heqin Zhu, Quan Quan, Qingsong Yao, Zaiyi Liu, and S Kevin Zhou. UOD: Universal one-shot detection of anatomical landmarks. In International Conference on Med- ical Image Computing and Computer-Assisted Intervention , pages 24–34. Springer, 2023. 1 CABLD: Contrast-Agnostic Bra...

    2023

  67. [75]

    Coordi- nates in D dimensions are represented as column vectors, i.e., x ∈ RD

    Analytical and Differentiable Coordinate Transformations Notation: Lowercase bold letters denote column vectors, while uppercase bold letters are used for matrices. Coordi- nates in D dimensions are represented as column vectors, i.e., x ∈ RD. The symbol ˜x denotes x in homoge...

  68. [76]

    The model consists of five non-linear blocks, each comprising an RC layer followed by a LeakyReLU activation

    Random Convolution-Based Contrast Aug- mentation Figure S1 illustrates the model architecture used for random convolution (RC)-based contrast augmentation. The model consists of five non-linear blocks, each comprising an RC layer followed by a LeakyReLU activation. This cascad...

  69. [77]

    This approach is based on the fact that RC does not alter the geometric properties of the scans but instead generates arbitrary contrast variations

    and the subsequent calculation of similarity and registra- tion loss functions. This approach is based on the fact that RC does not alter the geometric properties of the scans but instead generates arbitrary contrast variations. This forces the model to predict landmarks indep...

  70. [78]

    This outcome was expected, as 3D U-Net typically has a much heavier parameter load compared to simpler architectures like the 3D supervised CNN we implemented

    Baselines It is important to note that we did not include the 3D U- Net as one of our baselines for direct landmark detection because it failed to converge and performed poorly on the publicly available test sets. This outcome was expected, as 3D U-Net typically has a much hea...

  71. [79]

    S4 for comparison

    Visual Comparison with ANTs and Key- Morph Samples of landmarks generated from our proposed model, ANTs, and KeyMorph for the same subject are shown in the axial view in Fig. S4 for comparison. Note that all landmarks are in 3D. For easy visualization, we show the 3D points pr...

  72. [80]

    The resulting MREs (in mm) were 5.87±4.02, 5.56±3.51, 4.92±3.12, and 5.38±3.32 for the SNSX, OASIS, HCP, and HCP-T2w datasets, respec- tively

    Sensitivity to the Template Choice To assess the sensitivity of our method to the choice of tem- plate, we tested it using the widely adopted T1-weighted Colin27 atlas [21] (a young, single-subject template) as an extreme alternative. The resulting MREs (in mm) were 5.87±4.02,...

  73. [81]

    As an indirect accuracy test, we evaluated CABLD for Parkin- son’s disease (PD) and Alzheimer’s disease (AD) diagnosis (Sec

    Robustness to Pathological Brains While our current evaluation focuses on healthy subjects due to the availability of annotated data, assessing robust- ness for pathological brains is clinically important. As an indirect accuracy test, we evaluated CABLD for Parkin- son’s dise...

  74. [82]

    These demonstrate the computational efficiency of our framework for large- scale and time-sensitive applications

    Computation Time Our method achieves an average inference time of 0.35±0.012s (GPU) and 6.42±0.20s (CPU), which is significantly faster than ANTs (MI: 428.22±3.14s, CC: 380.62±0.72s, CPU), and also faster than KeyMorph (10.12±0.22s CPU, 0.54±0.01s GPU) and BrainMorph (180.14±1...

Pith tools

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