Pith. sign in

REVIEW 3 major objections 5 minor 34 references

Cascading 3D face-mesh models can classify clinical facial traits along the Human Phenotype Ontology rather than naming syndromes.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-11 05:24 UTC pith:KWRNGQ6H

load-bearing objection Solid hierarchical HPO phenotyper on MediaPipe meshes; parent nodes work, rare leaves do not, and the authors already document both. the 3 major comments →

arxiv 2607.05585 v1 pith:KWRNGQ6H submitted 2026-07-06 cs.CV cs.AIcs.LG

Hierarchical Classification via Cascading Feature Elimination: Application to Human Phenotype Ontology-Aligned Facial Phenotyping (FaceMesh2HPO)

Fabio Hellmann , Alexander Hustinx , Benjamin D. Solomon , GestaltMatcher Database Consortium , Tzung-Chien Hsieh , Peter Krawitz , Elisabeth Andr\'e This is my paper
classification cs.CV cs.AIcs.LG
keywords Face MeshFeature EliminationDeep LearningHuman Phenotype OntologySyndromesPointNetFacial PhenotypingHierarchical Classification
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Clinicians diagnose rare genetic conditions by recognizing structured facial traits, not by matching whole syndromes at once. This paper argues that the same reasoning can be automated by turning 2D clinical photographs into 3D face meshes and training a cascade of small geometric classifiers, one for each Human Phenotype Ontology term, that pass pruned landmark masks down the ontology tree. The resulting labels stay in clinical language, remain informative even when the underlying disorder is unseen, and can be inspected by region. Performance is strongest at parent and intermediate terms and weaker for rare, sparsely annotated leaves, so the method is presented as a phenotypic scaffold rather than a universal detector of every fine-grained trait.

Core claim

Geometric 3D face meshes combined with a hierarchical PointNet cascade that eliminates unimportant landmarks term by term can produce clinically meaningful, ontology-aligned facial phenotype predictions, with mean AUROC around 0.75 and systematically higher accuracy near the root than at rare leaves.

What carries the argument

Cascading feature elimination along the HPO tree: each PointNet classifier is trained on its subtree, Integrated Gradients ranks mesh points, a threshold prunes the mask, and the reduced mask is handed to child terms so that deeper models see only the landmarks that mattered higher up.

Load-bearing premise

That 478-point meshes extracted from ordinary 2D clinical photographs, together with incomplete clinician annotations and soft-labeled control faces, still carry enough reliable geometric signal to train the hierarchy.

What would settle it

Train and evaluate the same cascade on true 3D facial scans of the same patients and check whether leaf-term AUROCs rise substantially above chance and whether detection prevalence no longer drifts so far from the balanced training prior.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Diagnostic tools can output reusable HPO trait lists instead of a single syndrome name, supporting differential diagnosis and tools that already consume HPO terms.
  • Parent and compression nodes already give usable signal and transfer better to unseen disorders than leaf models, so coarser ontology levels are the practical entry point for clinical use.
  • Point-importance maps and expert region masks make the decision regions inspectable, giving a path to clinician-auditable geometric explanations.
  • The same cascade can be extended as more curated HPO annotations become available without redesigning the architecture.

Where Pith is reading between the lines

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

  • If mesh detectors systematically miss dysmorphic landmarks, the cascade will keep under-performing on the very traits clinicians care about most; refining the front-end mesh extractor may matter more than further architecture tweaks.
  • Soft labels and 50/50 rebalancing mask the true prevalence of rare traits, so any clinical deployment will need explicit calibration and prevalence-aware decision thresholds before the reported F1 numbers can be trusted at the bedside.
  • The large performance gap between parent and leaf terms suggests the method may be most valuable as a pre-filter that proposes intermediate HPO terms for human review rather than as an end-to-end leaf-term annotator.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. FaceMesh2HPO frames facial dysmorphism analysis as hierarchical HPO-term classification rather than syndrome-level prediction. From 2D clinical photographs the authors extract MediaPipe 478-point approximate-3D meshes, combine them with clinician-curated positive/negative HPO labels for 10 disorders (107 terms after parent expansion) plus soft-labeled UTKFace controls, and train a cascade of dynamically sized PointNets organized along a simplified HPO tree. After each node is trained, Integrated Gradients point importance is thresholded and the surviving mask is passed to child nodes (cascading feature elimination); optional expert region masks override the cascade for selected anatomical terms. A 72-configuration ablation selects 3D meshes + facial outline + age/sex/ethnicity metadata, threshold 0.01 and soft-label 0.05, yielding mean AUROC 0.750 ± 0.042; parent/compression nodes routinely exceed 0.85 while many rare leaves remain near chance. An independent multi-expert test set of seen and unseen disorders shows heterogeneous F1 transfer, stronger at coarser ontology levels. A local web tool is released for clinical use.

Significance. The work supplies a clinically aligned alternative to black-box syndrome classifiers: ontology-linked phenotype scores that can feed tools such as LIRICAL and remain informative for unseen disorders. Strengths include a large multi-clinician annotation effort (124 annotators), systematic ablation, stratified 5-fold CV that respects patient identity, explicit parent-versus-leaf and seen-versus-unseen reporting, and a privacy-preserving local web application. If the hierarchical mesh approach continues to improve with better landmark detectors and more diverse data, it could become a practical component of next-generation phenotyping pipelines. The contribution is incremental rather than transformative, but the empirical documentation of performance bounds is solid and useful for the community.

major comments (3)
  1. [§5.2 Results / Table 4 / Fig. 5] §5.2 / Table 4 / Fig. 5: leaf-term sample sizes frequently fall below 200 (e.g., Microphthalmia N=66, Blepharophimosis N=152) and the paper itself shows that AUROC ≥ 0.75 for leaves requires ≥ 988 samples. The central claim of “clinically meaningful classification” therefore holds mainly for parent/compression nodes; the manuscript should either restrict the primary claim to those levels or supply a quantitative decision rule (minimum N / minimum AUROC) for when a leaf model is considered usable.
  2. [§6 Discussion / citation [34]] §6 Discussion and the authors’ own citation [34]: MediaPipe meshes are known to under-represent traits such as Big Nose and Big Lips on dysmorphic faces. Because the entire cascade depends on these 478 points, the Discussion should quantify how often landmark failure occurs on the GMDB-HPO cohort (or at least report a failure-rate proxy) and discuss whether expert region masks can fully compensate. Without that evidence the geometric premise remains only partially validated.
  3. [§4.4 Training Strategy / Table 4] §4.4 / Table 4 detection-prevalence column: training forces a 50/50 class balance while detection prevalence ranges from ~0.23 to ~0.74. Soft labels of 0.05 mitigate but do not eliminate the resulting miscalibration. For clinical deployment the authors should report calibration metrics (Brier, ECE) per node on the external test set and, if necessary, apply the same beta/temperature calibration already used in the web tool to the offline evaluation tables.
minor comments (5)
  1. [§3.1 / Table 1] Table 1 and §3.1: two HPO terms (HP:0410030, HP:0004493) were dropped for insufficient data; state the exact sample-size cutoff used.
  2. [Fig. 2 / §4] Fig. 2 caption and §4: clarify that the displayed point-importance colors are Integrated Gradients aggregated across folds, not single-fold attributions.
  3. [Appendix Table A1] Appendix Table A1: several IRA values are reported as 1.00 ± 0.00; note whether these terms were annotated by only one rater or truly unanimous.
  4. [§4.3 Model-Architecture] §4.3: the dynamic width schedule (p = 6/5/4/3) is sensible but the zero-padding rule for <2 points should be stated more explicitly to aid re-implementation.
  5. [Tables / Fig. 6] Minor typos: “norphology” (Table C3), “unseed” (Fig. 6 caption), and inconsistent spelling of “softlabel/soft label”.

Circularity Check

0 steps flagged

No circularity: purely empirical hierarchical supervised learning whose reported AUROCs are measured against held-out folds and an external multi-expert set, not forced by construction or self-citation.

full rationale

FaceMesh2HPO trains a tree of PointNet classifiers on MediaPipe 478-point meshes extracted from 2D images, with cascading Integrated-Gradients point elimination along a simplified HPO hierarchy, soft-labeled UTKFace negatives, and optional demographic metadata. Every numerical claim (mean AUROC 0.750 ± 0.042, per-node ranges ≈0.55–0.89, parent-vs-leaf gaps, sample-size correlations, external F1 differences) is obtained by 5-fold stratified cross-validation or by evaluation on an independent multi-expert test set; none is algebraically identical to a fitted constant or to a definition. Ablations (Table 3 / B2) and per-term tables (C3, D4–D22) further document that performance varies with data support and ontology depth rather than being guaranteed by the architecture. Self-citations (GestaltMatcher, Hellmann et al. on MediaPipe fidelity) supply background data and acknowledged limitations; they do not underwrite the reported metrics. The derivation chain is therefore self-contained empirical machine learning with no circular reduction.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 2 invented entities

The central empirical claim rests on a handful of modeling choices (threshold, soft-label value, tree simplification) and domain assumptions about mesh fidelity and annotation quality; no new physical entities are postulated.

free parameters (4)
  • feature-importance elimination threshold = 0.01
    Chosen by ablation (0.01 best); directly controls how many landmarks survive for child models and therefore leaf-term performance.
  • soft-label value for negatives = 0.05
    Ad-hoc value (0.05) used because UTKFace and unannotated GMDB cases are not verified non-syndromic; selected by ablation.
  • PointNet width parameter p = 6/5/4/3 by point count
    Dynamically set by input point count (p=6/5/4/3); architectural free choice that affects capacity.
  • learning-rate / early-stopping patience = 0.0001 / 5
    Fixed at 1e-4 and patience 5; standard but still free hyper-parameters that influence reported AUROCs.
axioms (4)
  • ad hoc to paper HPO cross-links can be cut so that the remaining structure is a pure tree with Abnormality of the face as root.
    Required for the cascading parent-to-child mask-passing scheme; stated in Section 4.
  • domain assumption MediaPipe's 478-point mesh extracted from 2D clinical photographs is a sufficiently faithful geometric representation of dysmorphic facial morphology.
    Core input representation; authors note limitations for certain traits in their own prior work.
  • domain assumption Clinician annotations (present/absent/uncertain) plus soft-labeled UTKFace controls constitute usable ground truth for hierarchical binary classifiers.
    Training labels; incompleteness and inter-rater variability are acknowledged but treated as manageable.
  • domain assumption Integrated Gradients point-importance scores accumulated across folds yield a stable mask for cascading elimination.
    Mechanism that prunes landmarks for child models; no formal guarantee that clinically critical points survive.
invented entities (2)
  • Cascading feature-elimination hierarchy of PointNets aligned to HPO no independent evidence
    purpose: Progressively restrict mesh points so that leaf models see only landmarks deemed important by their ancestors.
    The core methodological invention; ablation shows it improves mean AUROC relative to non-cascaded baselines.
  • FaceMesh2HPO end-to-end pipeline and local web tool no independent evidence
    purpose: Package the trained models for clinical use with confidence calibration and importance visualization.
    Engineering contribution that makes the method usable; not independently validated outside the paper's own test set.

pith-pipeline@v1.1.0-grok45 · 60027 in / 2979 out tokens · 35112 ms · 2026-07-11T05:24:19.152440+00:00 · methodology

0 comments
read the original abstract

FaceMesh2HPO is a framework for classifying facial phenotypic descriptors aligned with the Human Phenotype Ontology (HPO) to support clinical diagnosis. Using annotations from 124 clinicians across 10 disorders (107 HPO terms) combined with non-syndromic controls, we generated 3D facial meshes (478 landmarks) from 2D images and trained a hierarchical PointNet-based pipeline with cascading classification and feature elimination. The best models, incorporating 3D meshes, facial outline, and demographic metadata, achieved AUROCs between ~0.55 and ~0.89, with higher performance at parent nodes than leaf terms. External validation showed variable generalizability across disorders. Results demonstrate that hierarchical modeling of 3D facial geometry enables interpretable, ontology-linked phenotype classification, though performance on rare leaf terms remains limited. Improved data diversity and feature selection strategies are needed to enhance robustness and clinical utility.

discussion (0)

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

Reference graph

Works this paper leans on

34 extracted references · 14 canonical work pages

  1. [1]

    Orphanet Journal of Rare Diseases17(1), 233 (2022) https://doi.org/10.1186/s13023-022-02358-x

    Bauskis, A., Strange, C., Molster, C., Fisher, C.: The diagnostic odyssey: insights 58 from parents of children living with an undiagnosed condition. Orphanet Journal of Rare Diseases17(1), 233 (2022) https://doi.org/10.1186/s13023-022-02358-x . Accessed 2026-06-07

    work page doi:10.1186/s13023-022-02358-x 2022

  2. [2]

    Genetics in Medicine23(2), 272–279 (2021) https://doi.org/10.1038/s41436-020-00975-0

    Michaels-Igbokwe, C., McInnes, B., MacDonald, K.V., Currie, G.R., Omar, F., Shewchuk, B., Bernier, F.P., Marshall, D.A.: (Un)standardized testing: the diag- nostic odyssey of children with rare genetic disorders in Alberta, Canada. Genetics in Medicine23(2), 272–279 (2021) https://doi.org/10.1038/s41436-020-00975-0 . Accessed 2026-06-07

    work page doi:10.1038/s41436-020-00975-0 2021

  3. [3]

    American Journal of Medical Genetics Part C: Seminars in Medical Genetics 193(3), 32061 (2023) https://doi.org/10.1002/ajmg.c.32061

    Hsieh, T., Krawitz, P.M.: Computational facial analysis for rare Mendelian disor- ders. American Journal of Medical Genetics Part C: Seminars in Medical Genetics 193(3), 32061 (2023) https://doi.org/10.1002/ajmg.c.32061

    work page doi:10.1002/ajmg.c.32061 2023

  4. [4]

    Nature Genetics54(3), 349–357 (2022) https://doi.org/10

    Hsieh, T.-C., Bar-Haim, A., Moosa, S., Ehmke, N., Gripp, K.W., Pantel, J.T., Danyel, M., Mensah, M.A., Horn, D., Rosnev, S., Fleischer, N., Bonini, G., Hustinx, A., Schmid, A., Knaus, A., Javanmardi, B., Klinkhammer, H., Les- mann, H., Sivalingam, S., Kamphans, T., Meiswinkel, W., Ebstein, F., Kr¨ uger, E., K¨ ury, S., B´ ezieau, S., Schmidt, A., Peters, ...

    2022

  5. [5]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pp

    Hustinx, A., Hellmann, F., S¨ umer, O., Javanmardi, B., Andr´ e, E., Krawitz, P., Hsieh, T.-C.: Improving deep facial phenotyping for ultra-rare disorder verifica- tion using model ensembles. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pp. 5018–5028 (2023)

    2023

  6. [6]

    Genetics in Medicine21(8), 1719–1725 (2019) https://doi.org/10.1038/ s41436-018-0404-y

    Van Der Donk, R., Jansen, S., Schuurs-Hoeijmakers, J.H.M., Koolen, D.A., Goltstein, L.C.M.J., Hoischen, A., Brunner, H.G., Kemmeren, P., Nell˚ aker, C., Vissers, L.E.L.M., De Vries, B.B.A., Hehir-Kwa, J.Y.: Next-generation pheno- typing using computer vision algorithms in rare genomic neurodevelopmental disorders. Genetics in Medicine21(8), 1719–1725 (201...

    2019

  7. [7]

    Nature Genetics55(9), 1598–1607 (2023) https://doi.org/10.1038/ s41588-023-01469-w

    Dingemans, A.J.M., Hinne, M., Truijen, K.M.G., Goltstein, L., Reeuwijk, J., Leeuw, N., Schuurs-Hoeijmakers, J., Pfundt, R., Diets, I.J., Hoed, J., Boer, E., Spek, J., Jansen, S., Bon, B.W., Jonis, N., Ockeloen, C.W., Vulto-van Silfhout, A.T., Kleefstra, T., Koolen, D.A., Campeau, P.M., Palmer, E.E., Van Esch, H., Lyon, G.J., Alkuraya, F.S., Rauch, A., Mar...

    2023

  8. [8]

    PLOS Genetics20(2), 1011168 (2024) https://doi.org/10.1371/journal.pgen.1011168

    Duong, D., Johny, A.R., Ledgister Hanchard, S., Fortney, C., Flaharty, K., Hell- mann, F., Hu, P., Javanmardi, B., Moosa, S., Patel, T., Persky, S., S¨ umer, O., Tekendo-Ngongang, C., Lesmann, H., Hsieh, T.-C., Waikel, R.L., Andr´ e, E., Krawitz, P., Solomon, B.D.: Comparison of clinical geneticist and computer visual attention in assessing genetic condit...

    work page doi:10.1371/journal.pgen.1011168 2024

  9. [9]

    The American Journal of Human Genetics83(5), 610–615 (2008) https://doi.org/10.1016/j.ajhg.2008.09.017

    Robinson, P.N., K¨ ohler, S., Bauer, S., Seelow, D., Horn, D., Mundlos, S.: The human phenotype ontology: A tool for annotating and analyzing human hered- itary disease. The American Journal of Human Genetics83(5), 610–615 (2008) https://doi.org/10.1016/j.ajhg.2008.09.017

    work page doi:10.1016/j.ajhg.2008.09.017 2008

  10. [10]

    iScience27(3), 109325 (2024) https://doi.org/10.1016/j.isci.2024.109325

    Qiao, H., Tan, J., Yan, J., Sun, C., Yin, X., Li, Z., Wu, J., Guan, H., Wen, S., Zhang, M., Xu, S., Jin, L.: A comprehensive evaluation of the phenotype-first and data-driven approaches in analyzing facial morphological traits. iScience27(3), 109325 (2024) https://doi.org/10.1016/j.isci.2024.109325

    work page doi:10.1016/j.isci.2024.109325 2024

  11. [11]

    Research Square, 3–4438861 (2024) https://doi.org/10.21203/rs.3.rs-4438861/v1

    Lesmann, H., Hustinx, A., Moosa, S., Klinkhammer, H., Marchi, E., Caro, P., Abdelrazek, I.M., Pantel, J.T., Hagen, M.t., Thong, M.-K., Binti Mazlan, R.A., Tae, S.K., Kamphans, T., Meiswinkel, W., Li, J.-M., Javanmardi, B., Knaus, A., Uwineza, A., Knopp, C., Tkemaladze, T., Elbracht, M., Mattern, L., Jamra, R.A., Velmans, C., Strehlow, V., Jacob, M., Peron...

    work page doi:10.21203/rs.3.rs-4438861/v1 2024

  12. [12]

    International Journal of Data Science and Analytics20(3), 2871–2889 (2025) https://doi.org/10.1007/ s41060-024-00658-y

    Islam, T.U., Shaikh, T.A.: A lightweight attention-based deep learning facial recognition system for multiple genetic syndromes. International Journal of Data Science and Analytics20(3), 2871–2889 (2025) https://doi.org/10.1007/ s41060-024-00658-y

    2025

  13. [13]

    Genetics in Medicine22(10), 1682–1693 (2020) https://doi.org/10.1038/ s41436-020-0845-y

    Hallgr´ ımsson, B., Aponte, J.D., Katz, D.C., Bannister, J.J., Riccardi, S.L., Mahasuwan, N., McInnes, B.L., Ferrara, T.M., Lipman, D.M., Neves, A.B., Spitz- macher, J.A.J., Larson, J.R., Bellus, G.A., Pham, A.M., Aboujaoude, E., Benke, T.A., Chatfield, K.C., Davis, S.M., Elias, E.R., Enzenauer, R.W., French, B.M., Pickler, L.L., Shieh, J.T.C., Slavotinek...

    2020

  14. [14]

    IEEE Journal of Biomedical and Health Informatics26(7), 3229–3239 (2022) https://doi.org/10.1109/JBHI

    Bannister, J.J., Wilms, M., Aponte, J.D., Katz, D.C., Klein, O.D., Bernier, F.P.J., Spritz, R.A., Hallgr´ ımsson, B., Forkert, N.D.: A Deep Invertible 3-D Facial Shape Model for Interpretable Genetic Syndrome Diagnosis. IEEE Journal of Biomedical and Health Informatics26(7), 3229–3239 (2022) https://doi.org/10.1109/JBHI. 2022.3164848

    doi:10.1109/jbhi 2022

  15. [15]

    IEEE Access10, 23450–23462 (2022) https://doi.org/10.1109/ ACCESS.2022.3153357

    Mahdi, S.S., Matthews, H., Nauwelaers, N., Vanneste, M., Gong, S., Bouritsas, G., Baynam, G.S., Hammond, P., Spritz, R., Klein, O.D., Hallgr´ ımsson, B., Peeters, H., Bronstein, M., Claes, P.: Multi-Scale Part-Based Syndrome Classification of 3D Facial Images. IEEE Access10, 23450–23462 (2022) https://doi.org/10.1109/ ACCESS.2022.3153357

    work page arXiv 2022

  16. [16]

    Dingemans, A.J.M., Vries, B.B.A.d., Vissers, L.E.L.M., Gerven, M.A.J.v., Hinne, M.: Comparing facial feature extraction methods in the diagnosis of rare genetic syndromes. medRxiv. Pages: 2022.08.26.22279217 (2022). https://doi.org/10. 1101/2022.08.26.22279217

    2022

  17. [17]

    American Journal of Human Genetics107(3), 403–417 (2020) https://doi.org/10.1016/j.ajhg.2020.06.021 61

    Robinson, P.N., Ravanmehr, V., Jacobsen, J.O.B., Danis, D., Zhang, X.A., Car- mody, L.C., Gargano, M.A., Thaxton, C.L., Karlebach, G., Reese, J., Holtgrewe, M., K¨ ohler, S., McMurry, J.A., Haendel, M.A., Smedley, D.: Interpretable Clin- ical Genomics with a Likelihood Ratio Paradigm. American Journal of Human Genetics107(3), 403–417 (2020) https://doi.or...

    work page doi:10.1016/j.ajhg.2020.06.021 2020

  18. [18]

    In: 2013 35th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp

    Ullah, M.Z., Aono, M., Seddiqui, M.H.: Estimating a ranked list of human heredi- tary diseases for clinical phenotypes by using weighted bipartite network. In: 2013 35th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp. 3475–3478 (2013). https://doi.org/10.1109/EMBC. 2013.6610290 . ISSN: 1558-4615

    doi:10.1109/embc 2013

  19. [19]

    Nucleic Acids Research33(Database issue), 514–517 (2004) https://doi.org/10.1093/nar/gki033

    Hamosh, A.: Online mendelian inheritance in man (omim), a knowledgebase of human genes and genetic disorders. Nucleic Acids Research33(Database issue), 514–517 (2004) https://doi.org/10.1093/nar/gki033

    work page doi:10.1093/nar/gki033 2004

  20. [20]

    Euro- pean Journal of Human Genetics28(2), 165–173 (2019) https://doi.org/10.1038/ s41431-019-0508-0

    Nguengang Wakap, S., Lambert, D.M., Olry, A., Rodwell, C., Gueydan, C., Lanneau, V., Murphy, D., Le Cam, Y., Rath, A.: Estimating cumulative point prevalence of rare diseases: analysis of the orphanet database. Euro- pean Journal of Human Genetics28(2), 165–173 (2019) https://doi.org/10.1038/ s41431-019-0508-0

    2019

  21. [21]

    https://www.orphadata

    INSERM: Orphadata: Free access data from Orphanet. https://www.orphadata. com/. INSERM, Orphanet. Available at: https://www.orphadata.com/. Data accessed: 22 April 2026 (1999)

    2026

  22. [22]

    Psychologi- cal Bulletin76(5), 378–382 (1971) https://doi.org/10.1037/h0031619

    Fleiss, J.L.: Measuring nominal scale agreement among many raters. Psychologi- cal Bulletin76(5), 378–382 (1971) https://doi.org/10.1037/h0031619 . Accessed 2026-06-30

    doi:10.1037/h0031619 1971

  23. [23]

    European Journal of Human Genetics33(3), 377–382 (2025) https: //doi.org/10.1038/s41431-025-01787-z

    Kirchhoff, A., Hustinx, A., Javanmardi, B., Hsieh, T.-C., Brand, F., Hellmann, F., Mertes, S., Andr´ e, E., Moosa, S., Schultz, T., Solomon, B.D., Krawitz, P.: Gestaltgan: synthetic photorealistic portraits of individuals with rare genetic disorders. European Journal of Human Genetics33(3), 377–382 (2025) https: //doi.org/10.1038/s41431-025-01787-z

    work page doi:10.1038/s41431-025-01787-z 2025

  24. [24]

    https://arxiv

    Kartynnik, Y., Ablavatski, A., Grishchenko, I., Grundmann, M.: Real-time Facial Surface Geometry from Monocular Video on Mobile GPUs (2019). https://arxiv. org/abs/1907.06724v1 Accessed 2026-04-08

    Pith/arXiv arXiv 2019

  25. [25]

    BioData Mining16(1) (2023) https://doi.org/10.1186/s13040-023-00322-4

    Chicco, D., Jurman, G.: The matthews correlation coefficient (mcc) should replace the roc auc as the standard metric for assessing binary classification. BioData Mining16(1) (2023) https://doi.org/10.1186/s13040-023-00322-4

    doi:10.1186/s13040-023-00322-4 2023

  26. [26]

    Qi, C.R., Su, H., Mo, K., Guibas, L.J.: PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation. arXiv. Version Number: 2 (2016). https: //doi.org/10.48550/ARXIV.1612.00593

    Pith/arXiv arXiv doi:10.48550/arxiv.1612.00593 2016

  27. [27]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2017)

    Zhang, Z., Song, Y., Qi, H.: Age progression/regression by conditional adversarial autoencoder. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2017). IEEE 62

    2017

  28. [28]

    In: Precup, D., Teh, Y.W

    Sundararajan, M., Taly, A., Yan, Q.: Axiomatic Attribution for Deep Net- works. In: Precup, D., Teh, Y.W. (eds.) Proceedings of the 34th Interna- tional Conference on Machine Learning, pp. 3319–3328. PMLR, ??? (2017). https://proceedings.mlr.press/v70/sundararajan17a.htmlAccessed 2026-06-07

    2017

  29. [29]

    Journal of the American Medical Informatics Association 21(3), 501–508 (2014) https://doi.org/10.1136/amiajnl-2013-001964

    Nguyen, Q., Valizadegan, H., Hauskrecht, M.: Learning classification models with soft-label information. Journal of the American Medical Informatics Association 21(3), 501–508 (2014) https://doi.org/10.1136/amiajnl-2013-001964

    work page doi:10.1136/amiajnl-2013-001964 2014

  30. [30]

    Biochimica et Biophysica Acta (BBA) - Protein Structure405(2), 442–451 (1975) https://doi.org/10.1016/0005-2795(75)90109-9

    Matthews, B.W.: Comparison of the predicted and observed secondary struc- ture of T4 phage lysozyme. Biochimica et Biophysica Acta (BBA) - Protein Structure405(2), 442–451 (1975) https://doi.org/10.1016/0005-2795(75)90109-9 . Accessed 2026-07-04

    doi:10.1016/0005-2795(75)90109-9 1975

  31. [31]

    In: Singh, A., Zhu, J

    Kull, M., Filho, T.S., Flach, P.: Beta calibration: a well-founded and easily imple- mented improvement on logistic calibration for binary classifiers. In: Singh, A., Zhu, J. (eds.) Proceedings of the 20th International Conference on Artificial Intel- ligence and Statistics. Proceedings of Machine Learning Research, vol. 54, pp. 623–631. PMLR, ??? (2017)....

    2017

  32. [32]

    Neural Computing and Applications36(14), 8073–8095 (2024) https://doi.org/10.1007/s00521-024-09505-4

    Balanya, S.A., Maro˜ nas, J., Ramos, D.: Adaptive temperature scaling for Robust calibration of deep neural networks. Neural Computing and Applications36(14), 8073–8095 (2024) https://doi.org/10.1007/s00521-024-09505-4 . Accessed 2026- 07-03

    work page doi:10.1007/s00521-024-09505-4 2024

  33. [33]

    Monthly Weather Review78(1), 1–3 (1950) https://doi.org/10

    Brier, G.W.: VERIFICATION OF FORECASTS EXPRESSED IN TERMS OF PROBABILITY. Monthly Weather Review78(1), 1–3 (1950) https://doi.org/10. 1175/1520-0493(1950)078⟨0001:VOFEIT⟩2.0.CO;2 . Accessed 2026-07-04

    1950

  34. [34]

    ACM Transactions on Multi- media Computing, Communications, and Applications21(1), 1–27 (2025) https: //doi.org/10.1145/3641107 63

    Hellmann, F., Mertes, S., Benouis, M., Hustinx, A., Hsieh, T.-C., Conati, C., Krawitz, P., Andr´ e, E.: GANonymization: A GAN-Based Face Anonymization Framework for Preserving Emotional Expressions. ACM Transactions on Multi- media Computing, Communications, and Applications21(1), 1–27 (2025) https: //doi.org/10.1145/3641107 63

    work page doi:10.1145/3641107 2025