CoChat
Sign Up
Cited by 44Medical Imaging and AnalysisMedical Image Segmentation TechniquesDental Radiography and Imaging

Learning to detect anatomical landmarks of the pelvis in X-rays from arbitrary views

Open access
Bier, Bastian, Goldmann, Florian, Zaech, Jan-Nico, Fotouhi, Javad, Hegeman, Rachel, Grupp, Robert, Armand, Mehran, Osgood, Greg +3 more·2019·International Journal of Computer Assisted Radiology and Surgery
DOI 10.1007/s11548-019-01975-5PMC PMC7297500
Open this paper in CoChat — free account

Abstract

Purpose Minimally invasive alternatives are now available for many complex surgeries. These approaches are enabled by the increasing availability of intra-operative image guidance. Yet, fluoroscopic X-rays suffer from projective transformation and thus cannot provide direct views onto anatomy. Surgeons could highly benefit from additional information, such as the anatomical landmark locations in the projections, to support intra-operative decision making. However, detecting landmarks is challenging since the viewing direction changes substantially between views leading to varying appearance of the same landmark. Therefore, and to the best of our knowledge, view-independent anatomical landmark detection has not been investigated yet.Methods In this work, we propose a novel approach to detect multiple anatomical landmarks in X-ray images from arbitrary viewing directions. To this end, a sequential prediction framework based on convolutional neural networks is employed to simultaneously regress all landmark locations. For training, synthetic X-rays are generated with a physically accurate forward model that allows direct application of the trained model to real X-ray images of the pelvis. View invariance is achieved via data augmentation by sampling viewing angles on a spherical segment of [Formula: see text].Results On synthetic data, a mean prediction error of 5.6 ± 4.5 mm is achieved. Further, we demonstrate that the trained model can be directly applied to real X-rays and show that these detections define correspondences to a respective CT volume, which allows for analytic estimation of the 11 degree of freedom projective mapping.Conclusion We present the first tool to detect anatomical landmarks in X-ray images independent of their viewing direction. Access to this information during surgery may benefit decision making and constitutes a first step toward global initialization of 2D/3D registration without the need of calibration. As such, the proposed concept has a strong prospect to facilitate and enhance applications and methods in the realm of image-guided surgery.

Integrity signals

Registry

Crossref · 10.1007/s11548-019-01975-5

Cites

32 of 32 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

2 of 7 cited sources checked · 1 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    The angular range used for training approximates the range of variation in X-ray images during surgical procedures on the pelvis.

    “This range approximates the range of variation in X-ray images during surgical procedures on the pelvis [13].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [13] Khurana B, Sheehan SE, Sodickson AD, Weaver MJ (2014) Pelvic ring fractures: what the orthopedic surgeon wants to know. Radiographics 34(5):1317–1333 source

      Two searches of Khurana et al.’s “Pelvic Ring Fractures: What the Orthopedic Surgeon Wants to Know” returned no full-text snippets, so the cited paper could not be checked directly. An unscoped search found nearly identical wording in “X-ray-transform Invariant Anatomical Landmark Detection for Pelvic Trauma Surgery” (score 0.457), where the sentence itself cites reference [14]. That match may be a secondary citation; it does not establish that Khurana et al. is the wrong source. (LLM, conf=0.65)

      Experimental — accuracy not yet measured
  • Findingown data

    On synthetic data, the mean prediction error across all projections and landmarks was 5.6 ± 4.5 mm (9.1 ± 7.4 pixels).

    “On synthetic data, a mean prediction error of 5.6 ± 4.5mm is achieved.”
    • Also stated in: § Synthetic data · plain
      “Quantitatively, the average distance between ground truth positions and detection across all projections and landmarks is 9.1 ± 7.4pixels (5.6 ± 4.5 mm).”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.
    • Rests on an assumption: The manually labeled 3D landmark positions in the CT volumes are accurate enough to serve as ground truth for evaluating detection.
  • Findingown data

    In real X-ray images, the automatic landmark detections agreed well with ground truth across various poses and truncations.

    “Overall, the well agreement between the automatic detections and the ground truth positions can be appreciated for various poses and truncations.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.
    • Rests on an assumption: The manually labeled 3D landmark positions in the CT volumes are accurate enough to serve as ground truth for evaluating detection.
  • Findingreasoning

    The trained model can be applied directly to real X-ray images without retraining.

    “Further, we demonstrate that the trained model can be directly applied to real X-rays”
    • Also stated in: § Data generation · plain
      “This allows direct application of the network model to real clinical X-ray images.”
    • Also stated in: § Experiments and results · plain
      “Note, that the network has not been re-trained for this purpose.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.
  • Findingown data

    The landmark detections on real X-rays define correspondences to a CT volume and allow analytic estimation of the 11 degree-of-freedom projective mapping.

    “show that these detections define correspondences to a respective CT volume, which allows for analytic estimation of the 11 degree of freedom projective mapping.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.
    • Rests on an assumption: The manually labeled 3D landmark positions in the CT volumes are accurate enough to serve as ground truth for evaluating detection.
  • May combine two claims. 15
  • Classified as probably not a substantial claim. 9

  • May combine two claims.

    Study design and conductcitation

    A physically accurate forward projector was used to generate realistic synthetic training data, allowing direct application of the network to real clinical X-ray images.

    “A realistic forward projector that accounts for physically accurate image formation, while being capable of fast data generation, was used to obtain realistic synthetic training data. This allows direct application of the network model to real clinical X-ray images.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [26] Unberath M, Zaech JN, Lee SC, Bier B, Fotouhi J, Armand M, Navab N (2018) Deepdrr–a catalyst for machine learning in fluoroscopy-guided procedures. In: International conference on medical image computing and computer-assisted intervention. Springer, New York source
      Experimental — accuracy not yet measured
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.

  • May combine two claims.

    Findingown data

    Some landmarks have lower average error (e.g., landmark #9, 5.26 pixels) while others are detected poorly (e.g., landmark #23, 26.05 pixels).

    “First, some landmarks have an overall lower error (e.g., land mark #9 with an average error of 5.26pixels), while others are detected poorly (e.g., landmark #23 with an average error of 26.05pixels).”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.
    • Rests on an assumption: The manually labeled 3D landmark positions in the CT volumes are accurate enough to serve as ground truth for evaluating detection.

  • May combine two claims.

    Findingown data

    Some landmarks can be detected equally well independent of viewing direction (e.g., landmark #11), while for others detectability varies across quadrants (e.g., landmark #19).

    “Third, some landmarks can be detected equally well, independently of the viewing direction (e.g., landmark #11), while for others, the detectability highly varies across the quadrants (e.g., landmark #19).”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.
    • Rests on an assumption: The manually labeled 3D landmark positions in the CT volumes are accurate enough to serve as ground truth for evaluating detection.

  • May combine two claims.

    Findingown data

    The reprojection error of the BB-based projection matrices was 2-5 pixels, which the authors consider appropriate as a reference for evaluating landmark detection.

    “The low reprojection error of 2–5pixel is in line with our expectations and suggests that the resulting projection matrices are an appropriate reference when evaluating the performance of our proposed view-invariant landmark detection.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S1 (asserted), S2 (checkable), S3 (load-bearing).

    Findingnone given

    The robustness toward unseen scenarios such as tools in the image or changes due to fractured anatomy must be improved.

    “First of all, the robustness toward unseen scenarios such as tools in the image or changes of the anatomy due to fractured anatomy must be improved.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.

  • May combine two claims.

    Findingown data

    The network was trained on synthetic data generated with a physics-based framework and evaluated on synthetic and real test sets, showing promising performance.

    “The network was trained on synthetic data generated using a physics-based framework and evaluated on both synthetic and real test sets, revealing promising performance.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Before this work, view-independent anatomical landmark detection in X-ray images had not been investigated.

    “Therefore, and to the best of our knowledge, view-independent anatomical landmark detection has not been investigated yet.”
    • Also stated in: § Introduction · hedged
      “To the best of our knowledge, there exists no approach that is able to detect anatomical landmarks in X-ray images independent of the viewing direction.”
    • Also stated in: § Discussion and conclusion · hedged
      “Due to the characteristics of transmission imaging, landmark appearances change substantially with the viewing direction making anatomical landmark detection a challenging task that has, to the best of our knowledge, not previously been addressed.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Access to landmark information during surgery may benefit decision making and is a first step toward global initialization of 2D/3D registration without calibration.

    “Access to this information during surgery may benefit decision making and constitutes a first step toward global initialization of 2D/3D registration without the need of calibration.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The 20,000 generated X-rays were split into training, validation, and testing such that images of one patient were not shared among sets.

    “In total, 20,000 X-rays were generated and split 18 × 1 × 1-fold into training, validation, and testing, where we ensured that images of one patient are not shared among these sets.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Manual labeling of landmarks is time-consuming, interrupts clinical workflow, and is subjective, yielding rater-dependent results.

    “Manual labeling is time-consuming, interrupts the clinical workflow, and is subjective, which yields rater dependent results.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [28] Wang CW, Huang CT, Hsieh MC, Li CH, Chang SW, Li WC, Vandaele R, Marée R, Jodogne S, Geurts P, Chen C, Zhen G, Chu C, Mirzaalian H, Vrtovec T, Ibragimov B (2015) Evaluation and comparison of anatomical landmark detection methods for cephalometric x-ray images: a grand challenge. IEEE Trans Med Imaging 34(9):1890–1900 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationnone given

    The proposed method is purely image-based and detects multiple anatomical landmarks independent of viewing direction.

    “In this work, we propose an automatic, purely image-based method to detect multiple anatomical landmarks in X-ray images independent of the viewing direction.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationnone given

    The closed-form solution for image pose estimation is simple but a maximum likelihood approach would likely yield superior results with statistical outliers.

    “In this work, we relied on a closed form solution to estimate the image pose which is compelling due to its simplicity, yet a more sophisticated approach based on maximum likelihood would certainly yield superior results in the presence of statistical outliers.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Considering the maximum belief is powerful for selecting reliably detected landmarks and can serve as a confidence measure.

    “In this task, we also showed that considering the maximum belief is powerful for selecting reliably detected landmarks. This additional information can be used as a confidence measure for further processing tasks.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The proposed concept has been transferred to projection images of knees to estimate involuntary motion during scans.

    “Recently, the proposed concept of view-invariant anatomical landmark detection has been transferred to projection images of knees in an attempt to estimate involuntary motion during scans [3].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [3] Bier B, Aschoff K, Syben C, Unberath M, Levenston M, Gold G, Fahrig R, Maier A (2018) Detecting anatomical landmarks for motion estimation in weight-bearing imaging of knees. In: International workshop on machine learning for medical image reconstruction. Springer, New York, pp 83–90 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Manual labeling of landmarks in the 3D volume is a possible source of error because it is inherently prone to errors.

    “A possible source of error might be introduced by the labeling of the landmarks in the 3D volume that, since manual, is inherently prone to errors.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    The network has a very large receptive field that increases over stages, enabling learning of spatial dependencies over long distances.

    “Furthermore, the network has a very large receptive field that also increases over stages, which enables the learning of spatial dependencies over long distances.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.

  • May combine two claims.

    Interpretationreasoning

    The method can identify corresponding landmarks between arbitrary views of the same anatomy and estimate pose relative to a pre-procedurally acquired volume without calibration.

    “Based on landmark knowledge, we can (a) identify corresponding landmarks between arbitrary views of the same anatomy and (b) estimate pose relative to a pre-procedurally acquired volume without the need for any calibration.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.
    • Rests on an assumption: The manually labeled 3D landmark positions in the CT volumes are accurate enough to serve as ground truth for evaluating detection.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    The lower accuracy at border views may be explained by more overlapping anatomy and less variation in training data sampled in that area.

    “This might be explained by the higher amount of overlapping anatomy from these directions as well as a lower amount of variation of the training data sampled in this area.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationnone given

    Computing the loss over all intermediate predictions counteracts vanishing gradients and guides the network to focus early on detection.

    “Lastly, the loss over all intermediate predictions btp is computed, which counteracts the vanishing gradient effect and simultaneously guides the network to focus early on the detection task.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The proposed concept has strong prospect to facilitate and enhance applications and methods in image-guided surgery.

    “As such, the proposed concept has a strong prospect to facilitate and enhance applications and methods in the realm of image-guided surgery.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.

  • May combine two claims.

    Interpretationown data

    The landmark detection accuracy was sufficient to generate DRRs very similar to target X-ray images, suggesting successful initialization of 2D/3D registration.

    “For these various poses, the landmark detection accuracy proves sufficient to achieve views that are very similar to the target X-ray image, suggesting successful initialization.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.
    • Rests on an assumption: The manually labeled 3D landmark positions in the CT volumes are accurate enough to serve as ground truth for evaluating detection.
21 of 21

  • The authors state this as an assumption.

    Anatomical landmarks exhibit strong constraints and characteristic patterns even when viewing angles are arbitrary.

    “Our assumption is that anatomical landmarks exhibit strong constraints and thus characteristic patterns even in the presence of arbitrary viewing angles.”

  • The paper relies on this without support.

    The manually labeled 3D landmark positions in the CT volumes are accurate enough to serve as ground truth for evaluating detection.

    “In total, 23 landmark positions were manually labeled in 20 CTs of male and female patients using 3D volume renderings in 3D Slicer [5].”
    “A possible source of error might be introduced by the labeling of the landmarks in the 3D volume that, since manual, is inherently prone to errors.”
2 of 2

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Unlike [16,19,20], the authors train networks on smaller datasets of actual fluoroscopy and achieve state-of-the-art results on clinically relevant data.

    “Contrary to [16,19,20], we train our networks using smaller datasets of actual fluoroscopy and achieve state-of-the-art results on clinically relevant data.”

    From Automatic Annotation of Hip Anatomy in Fluoroscopy for Robust and Efficient 2D/3D Registration (2020) · cites this paper as [19]

    Bier et al. (2019) say they trained on synthetic X-rays and then applied the model to real X-rays, contrary to the claim that they trained on actual fluoroscopy. The quoted sentence appears instead in “Automatic Annotation of Hip Anatomy in Fluoroscopy for Robust and Efficient 2D/3D Registration” (unscoped score 0.52). That paper explicitly contrasts its approach with Bier’s, which it cites as [19]; it does not attribute its own training method to Bier. (LLM, conf=0.58)

    Checked by a reasoning model — may be wrong.

    Experimental — accuracy not yet measured

  • Bier trained CNNs to annotate pelvic anatomical landmarks using simulated fluoroscopy; on five sequences of actual fluoroscopy from two cadaveric specimens, mean annotation errors of 12-24 mm in the detector plane were reported.

    “Using a large collection of simulated fluoroscopy, Bier trained CNNs to annotate anatomical landmarks of the pelvis [19]. When evaluated on five sequences of actual fluoroscopy across two cadaveric specimens, mean annotation errors of 12-24 mm in the detector plane were reported.”

    From Automatic Annotation of Hip Anatomy in Fluoroscopy for Robust and Efficient 2D/3D Registration (2020) · cites this paper as [19]

    Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

    Experimental — accuracy not yet measured

  • Pelvis poses estimated from Bier's annotations produced reprojection errors of 14-34 mm for other landmarks not learned by the network.

    “Pelvis poses were estimated using these annotations, yielding reprojection errors of 14-34 mm for other landmarks not learned by the network.”

    From Automatic Annotation of Hip Anatomy in Fluoroscopy for Robust and Efficient 2D/3D Registration (2020) · cites this paper as [19]

    The Materials and methods section (‘2D/3D registration’; page unavailable) says detected pelvis landmarks were matched to CT labels to estimate a projection matrix, supporting the pose-estimation part. The Clinical data results (page unavailable) do not report 14–34 mm reprojection errors for landmarks the network did not learn; their stated 2–5 pixel error concerns a different, metal-bead-based reference projection. (LLM, conf=0.87)

    Experimental — accuracy not yet measured

  • Feature-based 2D/3D registration methods need to define meaningful landmark features manually and do not learn registration in an end-to-end fashion.

    “However, these methods need to define meaningful landmark features manually and do not learn registration in an end-to-end fashion.”

    From A Fully Differentiable Framework for 2D/3D Registration and the Projective Spatial Transformers (2024) · cites this paper as [16]

    In Materials and methods (Data generation), the authors say they manually selected clinically meaningful landmarks. In Materials and methods (2D/3D registration), they describe learned 2D landmark detection followed by a separate closed-form registration calculation, rather than end-to-end learning of registration. This supports the claim for the paper’s method, but not its implication that feature-based methods generally require manually defined landmarks. (LLM, conf=0.85)

    Experimental — accuracy not yet measured

  • There are popular feature-based 2D/3D registration methods, such as solving a PnP problem using corresponding anatomical landmarks, which may be complementary to intensity-based registration methods.

    “We are aware that there are popular feature-based 2D/3D registration methods, such as solving a PnP problem using corresponding anatomical landmarks [13], [16], [17], which may be complementary to the intensity-based registration methods.”

    From A Fully Differentiable Framework for 2D/3D Registration and the Projective Spatial Transformers (2024) · cites this paper as [16]

    The Materials and methods section (2D/3D registration; page unavailable) describes using corresponding 2D X-ray and 3D CT landmarks to initialize registration by estimating a projection matrix. It specifies a direct linear transform, not a PnP solver, and does not explicitly discuss complementarity with intensity-based registration. The Discussion and conclusion likewise describes landmark-based initialization, supporting that part of the claim. (LLM, conf=0.85)

    Experimental — accuracy not yet measured

Cites

Flagged first, then by citations

2 of the 34 references couldn’t be matched to a known paper.

Cited by

Sorted by citations

Related

Authors
Bier, B., Goldmann, F., Zaech, J.-N., Fotouhi, J., Hegeman, R., Grupp, R., Armand, M., Osgood, G., Navab, N., Maier, A., & Unberath, M.
Affiliations
  • Johns Hopkins University (US)
  • Friedrich-Alexander-Universität Erlangen-Nürnberg (DE)
Corresponding author
Bier, Bastian
ORCID
Venue
International Journal of Computer Assisted Radiology and Surgery
Publisher
Springer Science and Business Media LLC
Published
2019-04-20
Volume
14
Issue
9
Pages
1463-1473
ISSN
  • 1861-6410 (ISSN-L)
  • 1861-6429
DOI
10.1007/s11548-019-01975-5
PMC
PMC7297500
OpenAlex
W2936692422
Cited by
43
References
34
Citations by year
2026: 4 · 2025: 6 · 2024: 4 · 2023: 9 · 2022: 7 · 2021: 4 · 2020: 7 · 2019: 2
FWCI
1.76
Type
journal article
Language
en
Open access
Licence
Locations
  • International Journal of Computer Assisted Radiology and Surgery (publishedVersion)
  • PubMed (publishedVersion)
  • PubMed Central (acceptedVersion)
Funders
  • National Institutes of Health (R01 EB023939, R21 EB020113, R01 EB016703, R01 EB0223939, 5R01AR065248-03)
  • Nvidia (GPU donation)
Topics
  • Physical Sciences › Engineering › Biomedical Engineering › Medical Imaging and Analysis
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Medical Image Segmentation Techniques
  • Health Sciences › Dentistry › Oral Surgery › Dental Radiography and Imaging
Keywords
anatomical landmark detection, pelvis, fluoroscopy, homography, 2D-3D registration, data augmentation, minimally invasive surgery
MeSH terms
Algorithms, Calibration, Female, Humans, Male, Pelvis, Radiography, Tomography, X-Ray Computed, X-Rays, Reproducibility of Results, Neural Networks, Computer, Imaging, Three-Dimensional, Surgery, Computer-Assisted

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-09-29.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-09-29.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review