Pith. sign in

REVIEW 3 major objections 5 minor 48 references

A fully automated four-step pipeline regresses landmarks from fetal brain MRI label maps and refines them geometrically, achieving mean absolute errors below 2 mm for most of five biometric measurements.

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 →

A four-step deep learning pipeline automatically measures five fetal brain parameters from 3D MRI, with landmark-based interpretability, achieving sub-2 mm errors on most measurements but a 4.5 mm error on corpus callosum length.

T0 review reviewed 2026-08-05 challenge →

load-bearing objection Solid, honest pipeline for fetal MRI biometry, but the LCC proxy bias and an overstating abstract keep it from earning 'reliable biometry'. the 3 major comments →

arxiv 2608.03724 v1 pith:VJDERTNT submitted 2026-08-04 cs.CV

Towards Reliable and Reproducible Fetal Brain Biometry: A Deep Learning Approach Using MRI

classification cs.CV
keywords fetal brain MRIbiometrylandmark localizationdeep learninggeometric optimizationcorpus callosumreproducibility3D super-resolution
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.

The reading

This paper claims that fetal brain MRI biometry, normally a manual and time-consuming measurement task, can be automated end to end with enough accuracy for clinical use. The proposed four-step pipeline learns to place pairs of anatomical landmarks directly from brain tissue segmentation maps, then uses a measurement-specific geometric optimization to refine the landmark positions and read off distances. Across 150 volumes from two public fetal MRI datasets spanning 20–37 weeks of gestation, most of the five studied biometrics are measured with mean absolute errors below 2 mm and landmark localization errors mostly below 4 mm. The one clear exception is corpus callosum length, which is systematically underestimated despite a dedicated optimization, a limitation the authors report openly. If the central claim holds, automated landmark-based biometry could relieve radiologists of slow manual measurements while giving clinicians visible landmark positions to check.

Core claim

The central claim is that a straightforward, fully automated pipeline can jointly estimate linear biometric measurements and the anatomical landmarks that define them from 3D reconstructed fetal brain MRI. A 3D convolutional neural network regresses coarse landmark coordinates from tissue label maps; a measurement-specific geometric refinement step then projects each landmark pair onto the relevant anatomical structure and chooses the pair with maximum separation along the measurement axis. On five clinically used measurements—corpus callosum length, vermis height, transverse cerebellar diameter, brain biparietal diameter, and skull biparietal diameter—the method achieves mean absolute error

What carries the argument

The load-bearing object is the label map: a 3D volume in which every voxel is assigned a tissue class. A 3D CNN regresses landmark coordinates from these label maps, and a measurement-specific geometric optimizer refines each pair by defining the measurement axis, sampling points along it, keeping only points inside the relevant tissue mask, and selecting the two points with the largest Euclidean separation. For corpus callosum length, where the segmentation has no dedicated corpus-callosum label, the optimizer uses the white-matter mask as a proxy and snaps landmarks to white-matter/gray-matter and white-matter/cerebrospinal-fluid interfaces. This division of labor—CNN for initialization, t

Load-bearing premise

The fragile premise is that, for corpus callosum length, the boundary of the white-matter mask is a faithful proxy for the genu and splenium of the corpus callosum; the paper's own results show that this is the one measurement with significant systematic underestimation.

What would settle it

Run the released pipeline on a held-out set of fetal brain MRI volumes with landmarks labeled independently by two radiologists; if any measurement's mean absolute error exceeds the inter-rater difference, the claim that the pipeline matches manual variability fails. For corpus callosum specifically, retraining with a dedicated corpus-callosum label and still seeing MAE above 4 mm would falsify the paper's proxy-based explanation of its main failure.

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

If this is right

  • Fully automatic fetal brain biometry can be built without hand-crafted slice selection or direct regression of measurement values: landmark initialization plus geometric refinement is enough.
  • Because the pipeline consumes label maps rather than raw intensities, it can be extended to new measurements by adding or adjusting tissue labels, without retraining the whole system.
  • Outputting landmark positions alongside measurements gives clinicians an interpretable check: a questionable value can be verified by viewing the inferred landmarks in 3D.
  • The reported costs—about an hour of training on a mid-range GPU and seconds per subject at inference—make routine clinical integration feasible.
  • Corpus callosum length is the bottleneck; the paper's analysis points to a dedicated corpus-callosum label or better contrast as the concrete next step.

Where Pith is reading between the lines

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

  • The most direct test of the paper's explanation for the corpus-callosum bias is to add an explicit corpus-callosum class to the segmentation; if the systematic underestimation disappears, the white-matter proxy was the culprit.
  • The axis-wise landmark error pattern suggests that for the two biparietal diameters the limiting factor is identifying the correct imaging plane, not in-plane localization; a plane-normalization step could transfer to other volumetric landmark tasks.
  • The geometric optimizer's 'maximize distance inside a mask' rule implicitly assumes measurements span the extreme boundary of a healthy structure; malformations that distort boundaries are the most likely setting where the rule would need modification.
  • The same label-map-plus-geometric-refinement recipe could be tested on other modalities, such as ultrasound, where landmark-based fetal biometry is standard but manual measurement variability remains a problem.
Share X Bluesky LinkedIn Reddit HN

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. The manuscript presents a fully automated pipeline for fetal brain MRI biometry. A 3D CNN regresses landmark coordinates from 8-tissue label maps derived from T2-weighted super-resolution reconstructions; a measurement-specific geometric optimization refines the landmarks, and Euclidean distances between refined landmark pairs yield five biometric parameters (LCC, HV, TCD, bBIP, sBIP). The pipeline is trained and evaluated on 150 volumes from the Zurich and dHCP datasets with an 80/10/10 split, and it is compared with the publicly available pipeline by Luis et al. The authors report excellent inter-rater ICC values, low MAE for most measurements, landmark localization accuracy, and code/docker release.

Significance. If the evaluation supported the claims, the paper would make a useful contribution: it provides a public implementation, a systematic landmark-level evaluation, and an analysis of sensitivity to label-map source (GT vs BOUNTI). However, the central reliability claim is currently undercut by two load-bearing issues: the abstract's landmark-error claim is contradicted by the paper's own Table 5, and the one measurement with a dedicated optimization (LCC) shows the worst accuracy and a large systematic bias relative to the competitor. The authors are transparent about LCC being challenging, but that limitation directly affects the headline claim of reliable biometry for a clinically important parameter.

major comments (3)
  1. [Abstract; Table 5] The abstract states that the framework achieves 'mean landmark localization errors below 4 mm.' Table 5 reports mean landmark errors of 4.04±3.08 and 4.26±2.56 mm for LCC, 6.00±3.69 and 4.89±3.07 mm for bBIP, and 5.42±3.07 and 4.62±1.89 mm for sBIP. The claim is therefore false as written. The text in Section 3.3 correctly says 'most landmarks achieve a median Euclidean error below 4 mm,' but the abstract overstates this. Please revise the abstract to report median errors or to state that some landmarks exceed 4 mm on average.
  2. [Section 2.2.3; Table 4; Figure 6] The corpus callosum optimization uses the whole WM mask as a proxy for the corpus callosum and snaps landmarks to WM-GM and WM-CSF interfaces. Anatomically, the genu and splenium are not necessarily located at the outer boundary of the entire WM compartment, so this procedure can systematically select incorrect interface points and shorten LCC. The reported results are consistent with this concern: LCC is the only measurement with a significant paired difference from ground truth (p=0.0003 overall; Zurich p=5.98e-5; dHCP p=0.002), and the only measurement where the proposed method is substantially worse than Luis et al. (MAE 4.47 mm vs 1.77 mm; RMSE 5.45 vs 2.15; Cliff's delta=0.843 in favor of the competitor). Because LCC is clinically important for detecting callosal anomalies, the central claim of reliable biometry is conditional on an unvalidated proxy. I recommend either adding a de
  3. [Section 2.2.2; Section 3.2] The evaluation uses a fixed 80/10/10 split of 150 volumes, giving a test set of about 15 subjects, with no cross-validation and no data augmentation. After stratification by dataset, the per-dataset test subsets are roughly 7-8 subjects each. The claim that performance is 'consistent across datasets' is therefore based on very small samples. Please report the exact test-set sizes for each dataset, provide confidence intervals for the metrics, and temper the generalizability claims. The fixed split and absence of cross-validation are acknowledged, but they are load-bearing for the 'reliable and reproducible' claim.
minor comments (5)
  1. [Section 3.2] The text says 'Table 5 summarizes error metrics (MAE and RMSE) and R for our results,' but the MAE/RMSE/R values are in Table 4; Table 5 reports landmark localization errors. Please correct the table reference.
  2. [Section 3.2] The text cites 'Fig. 4' when discussing error distributions; the relevant figure appears to be Fig. 5.
  3. [Section 4] In the discussion, LCC is described as showing 'the lowest ICC in manual annotation (Table 4),' but the ICC values are in Table 3, not Table 4.
  4. [Table 4 caption] The caption states that the second and third rows report MAE and RMSE, and then says the third row includes Pearson R. This is confusing; please renumber the rows explicitly.
  5. [Supplementary S2] The dilation parameter k was lowered from 5 to 3 and 'selected as the optimal parameter for all subsequent analyses.' Please clarify whether this selection was made on the training/validation partition or on the test set; if the test set was used for tuning, this should be disclosed and the evaluation adjusted.

Circularity Check

0 steps flagged

No significant circularity: the pipeline is a supervised landmark-regression method evaluated on a held-out test split against manual ground truth and an external benchmark; the LCC proxy issue is a validity limitation, not a circular derivation.

full rationale

The paper's derivation chain is: segmentation label maps -> 3D CNN landmark regression -> measurement-specific geometric optimization -> Euclidean-distance measurements. The CNN is trained on manual landmark annotations and evaluated on a held-out test set (80/10/10 split stratified by dataset and gestational age), with all accuracy metrics computed against independent manual ground-truth landmarks. No fitted parameter or optimization objective is defined in terms of the test measurements or the final reported errors; the landmark refinement uses only anatomical masks and interface constraints. The LCC measurement uses a white-matter mask as a proxy for the corpus callosum, which is an anatomical approximation that plausibly explains the observed systematic LCC underestimation (MAE 4.47 mm vs 1.77 mm for the competitor), but this is a measurement-validity and accuracy limitation, not circularity: the proxy is not derived from the ground-truth LCC values, and the paper does not claim LCC is predicted from its own ground truth. The only self-citations are methodological (CNN architecture adapted from the authors' prior radiotherapy work [35]; super-resolution reliability from the authors' earlier fetal MRI study [13]), and they are not load-bearing justifications of the biometry results; the method is independently benchmarked against the external Luis et al. pipeline on two public datasets. Thus no step in the claimed derivation reduces by construction to its own input.

Axiom & Free-Parameter Ledger

2 free parameters · 4 axioms · 0 invented entities

The key assumptions are that segmentation labels faithfully represent anatomical structures, that a single 36-week atlas is a valid registration target across the whole 20-37 week range, that the manual dHCP annotations are reliable, and that the fixed split without augmentation yields a representative test set. One hand-tuned parameter (k dilation size) and a chosen neighborhood size for LCC are free parameters.

free parameters (2)
  • k landmark dilation size = 3 (minimum preserving all landmarks)
    Chosen heuristically (start 5) and reduced to minimum value ensuring landmark preservation across registration; affects preprocessing and potentially landmark retention.
  • LCC mid-sagittal neighborhood size = 5 slices (offsets -2 to +2)
    Hand-chosen to account for landmark variability; affects LCC optimization.
axioms (4)
  • domain assumption The input label maps (manual or BOUNTI) correctly delineate the anatomical structures used for landmark localization and geometric optimization.
    Section 2.1.2 and 2.2.3: the optimization restricts landmark candidates to specific labels (GM, WM, CSF, skull), so inaccurate segmentations would misplace landmarks; LCC error is attributed to missing CC label.
  • domain assumption The Uus 36-week atlas is an appropriate registration target for fetuses aged 20-37 weeks.
    Section 2.2.1: all volumes are rigidly registered to the Uus atlas at 36 weeks; large age span may strain this single target.
  • domain assumption Manual landmark annotations on dHCP, performed by two trained raters under supervision, are reliable ground truth.
    Section 2.1.3: dHCP lacked landmarks, so the authors annotated them; ICC was high (above 0.92) on a 10-subject subset.
  • ad hoc to paper No data augmentation and a fixed random split (80/10/10) yield a representative test set.
    Section 2.2.2: no augmentation is used; split preserves GA proportions, but test set is only about 15 subjects.

reviewed 2026-08-05 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Reliable and Reproducible Fetal Brain Biometry: A Deep Learning Approach Using MRI." pith.science (2026). https://pith.science/paper/VJDERTNT

@misc{pith2026260803724,
  author       = {Pith},
  title        = {Pith review of: Towards Reliable and Reproducible Fetal Brain Biometry: A Deep Learning Approach Using MRI},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VJDERTNT}},
  note         = {Machine review of arXiv:2608.03724}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Fetal brain biometry is essential for quantitative assessment of brain development, supporting gestational age estimation, developmental monitoring, and detection of abnormalities. In clinical practice, measurements are manually performed, making them time-consuming and prone to variability. While automated approaches have been proposed, reproducible methods remain limited, particularly those providing anatomically interpretable landmark localization. We present a fully automated deep learning-based framework for reliable and reproducible brain biometry from 3D super-resolution-reconstructed fetal brain MRI. The proposed four-step pipeline derives biometric parameters by jointly estimating linear measurements and their corresponding anatomical landmarks. A 3D convolutional neural network is trained to regress landmark coordinates from brain segmentation label maps, followed by measurement-specific geometric optimization to refine landmark positions and compute measurements. The pipeline is evaluated on two publicly available fetal MRI datasets comprising 150 volumes (gestational age range: 20-37 weeks) acquired across different scanners and protocols, assessing five key biometric measurements across varying acquisition settings and providing a comprehensive evaluation of both measurement accuracy and landmark localization using quantitative metrics and visual assessment. Compared with the only available automated pipeline, the proposed method achieves comparable or improved accuracy for most measurements. In conclusion, we introduce a straightforward pipeline for reliable biometry estimations, with efficiency, interpretability and scalability that support integration into clinical workflows.

Figures

Figures reproduced from arXiv: 2608.03724 by Denis Peruzzo, Francesca Maccarone, Giorgio Longari, Giulia Frigerio, Gloria Rizzato, Marina Di Stefano, Nivedita Agarwal, Rocco Prudentino, Simone Melzi, Tommaso Ciceri.

Figure 1
Figure 1. Figure 1: Distribution of subjects across gestational ages for the two datasets: Zurich and dHCP. For each GA, bars represent the cumulative number of subjects in both datasets splitted into subjects from the Zurich dataset (blue) and from the dHCP [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Example of the selected five biometrical measures. For each measure, corresponding landmarks manually placed on a 27-week GA subject from the Zurich dataset are shown. Panel (a) reports HV (green) and LCC (red) in sagittal view; panel (b) reports TCD (orange) in coronal view; panel (c) and (d) report bBIP (cyan) and sBIP (magenta), respectively, in the axial plane. bBIP: biparietal brain diameter; HV: heig… view at source ↗
Figure 5
Figure 5. Figure 5: Error distributions across test set subjects. The top row reports the absolute and percentage error distributions for each biometric measurement. The bottom row reports the absolute and percentage error distributions subdivided according to the datasets: for each measure, the left box refers to the Zurich dataset, while the right to the dHCP dataset [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: LCC results. Absolute error (in millimetres), and percentage error distributions in the test set for the LCC measurement with (LCC_opt) and without (LCC_noOpt) the application of the ad-hoc optimization process to the measurement landmarks (see ‘Corpus callosum optimization’ in subsection 2.2.3). Acronym: Opt= optimization 3.3. Landmarks analysis We complement the measurement error analysis by evaluating t… view at source ↗

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

48 extracted references · 37 canonical work pages · 1 internal anchor

  1. [1]

    Fetal biometry to assess the size and growth of the fetus

    O’Gorman, N., Salomon, L.J., 2018. Fetal biometry to assess the size and growth of the fetus. Best Pract. Res. Clin. Obstet. Gynaecol. 49, 3–15. https://doi.org/10.1016/j.bpobgyn.2018.02.005

  2. [2]

    Automatic biometry of fetal brain MRIs using deep and machine learning techniques

    She, J., Huang, H., Ye, Z., Huang, W., Sun, Y., Liu, C., et al., 2023. Automatic biometry of fetal brain MRIs using deep and machine learning techniques. Sci. Rep. 13, 17860. https://doi.org/10.1038/s41598 -023-43867-4

  3. [3]

    Towards automated fetal brain biometry reporting for 3-dimensional T2-weighted 0.55–3T magnetic resonance imaging at 20–40 weeks gestational age range

    Luis, A., Uus, A., Matthew, J., Arulkumaran, S., Egloff Collado, A., Kyriakopoulou, V., et al., 2025. Towards automated fetal brain biometry reporting for 3-dimensional T2-weighted 0.55–3T magnetic resonance imaging at 20–40 weeks gestational age range. Pediatr. Radiol. https://doi.org/10.1007/s00247-025-06403-2

  4. [4]

    Automatic linear measurements of the fetal brain on MRI with deep neural networks

    Avisdris, N., Yehuda, B., Ben-Zvi, O., Link-Sourani, D., Ben-Sira, L., Miller, E., et al., 2021. Automatic linear measurements of the fetal brain on MRI with deep neural networks. Int. J. Comput. Assist. Radiol. Surg. 16, 1481–92. https://doi.org/10.1007/s11548-021-02436-8

  5. [5]

    Corpus callosum length by gestational age as evaluated by fetal MR imaging

    Harreld, J.H., Bhor, R., Chason, D.P., Twickler, D.M., 2011. Corpus callosum length by gestational age as evaluated by fetal MR imaging. AJNR Am. J. Neuroradiol. 32, 490–4. https://doi.org/10.3174/ajnr.A2310

  6. [6]

    Artificial intelligence in fetal brain imaging: advancements, challenges, and multimodal approaches for biometric and structural analysis

    Wang, L., Fatemi, M., Alizad, A., 2025. Artificial intelligence in fetal brain imaging: advancements, challenges, and multimodal approaches for biometric and structural analysis. Comput. Biol. Med. 192, 110312. https://doi.org/10.1016/j.compbiomed.2025.110312

  7. [7]

    Normative linear and volumetric biometric measurements of fetal brain development in magnetic resonance imaging

    Cai, S., Zhang, G., Zhang, H., Wang, J., 2020. Normative linear and volumetric biometric measurements of fetal brain development in magnetic resonance imaging. Child’s Nerv. Syst. 36, 2997 –3005. https://doi.org/10.1007/s00381-020- 04633-3

  8. [8]

    Fetal central nervous system biometry on MR imaging

    Reichel, T.F., Ramus, R.M., Caire, J.T., Hynan, L.S., Magee, K.P., Twickler, D.M., 2003. Fetal central nervous system biometry on MR imaging. AJR Am. J. Roentgenol. 180, 1155–8. https://doi.org/10.2214/ajr.180.4.1801155

  9. [9]

    Motion - compensation techniques in neonatal and fetal MR imaging

    Malamateniou, C., Malik, S.J., Counsell, S.J., Allsop, J.M., McGuinness, A.K., Hayat, T., et al., 2013. Motion - compensation techniques in neonatal and fetal MR imaging. AJNR Am. J. Neuroradiol. 34, 1124–36. https://doi.org/10.3174/ajnr.A3128

  10. [10]

    Fetal MRI: what’s new? A short review

    Manganaro, L., Capuani, S., Gennarini, M., Miceli, V., Ninkova, R., Balba, I., et al., 2023. Fetal MRI: what’s new? A short review. Eur. Radiol. Exp. 7, 41. https://doi.org/10.1186/s41747 -023-00358-5

  11. [11]

    Review of deep learning and artificial intelligence models in fetal brain magnetic resonance imaging

    Vahedifard, F., Adepoju, J.O., Supanich, M., Ai, H.A., Liu, X., Kocak, M., et al., 2023. Review of deep learning and artificial intelligence models in fetal brain magnetic resonance imaging. World J. Clin. Cases 11, 3725 –35. https://doi.org/10.12998/wjcc.v11.i16.3725

  12. [12]

    Enhancing prenatal diagnosis: automated fetal brain MRI morphometry

    Masterl, E., Parkelj, A., Vesnaver, T.V., Špiclin, Ž., 2025. Enhancing prenatal diagnosis: automated fetal brain MRI morphometry. In: Link-Sourani, D., Abaci Turk, E., Macgowan, C., Hutter, J., Melbourne, A., Licandro, R. (Eds.), Perinatal, Preterm and Paediatric Image Analysis. PIPPI 2024. Lecture Notes in Computer Science. Springer, Cham, p. 14747:10. h...

  13. [13]

    Geometric reliability of super- resolution reconstructed images from clinical fetal MRI in the second trimester

    Ciceri, T., Squarcina, L., Pigoni, A., Ferro, A., Montano, F., Bertoldo, A., et al., 2023. Geometric reliability of super- resolution reconstructed images from clinical fetal MRI in the second trimester. Neuroinformatics 21, 549–63. https://doi.org/10.1007/s12021-023-09635-5

  14. [14]

    Deep learning model for predicting gestational age after the first trimester using fetal MRI

    Kojita, Y., Matsuo, H., Kanda, T., Nishio, M., Sofue, K., Nogami, M., et al., 2021. Deep learning model for predicting gestational age after the first trimester using fetal MRI. Eur. Radiol. 31, 3775 –82. https://doi.org/10.1007/s00330-021- 07915-9

  15. [15]

    Automatic ventriculomegaly detection in fetal brain MRI: a step-by-step deep learning model for novel 2D–3D linear measurements

    Vahedifard, F., Ai, H.A., Supanich, M.P., Marathu, K.K., Liu, X., Kocak, M., et al., 2023. Automatic ventriculomegaly detection in fetal brain MRI: a step-by-step deep learning model for novel 2D–3D linear measurements. Diagnostics 13,

  16. [16]

    Fetal gestational age prediction in brain MRI using artificial intelligence: a comparative study of three biometric techniques

    Vahedifard, F., Liu, X., Marathu, K.K., Ai, H.A., Supanich, M.P., Kocak, M., et al., 2024. Fetal gestational age prediction in brain MRI using artificial intelligence: a comparative study of three biometric techniques. Reprod. Med. 5, 113–35. https://doi.org/10.3390/reprodmed5030012

  17. [17]

    Fetal brain MRI measurements using a deep learning landmark network with reliability estimation, in: Sudre, C.H., et al

    Avisdris, N., Ben Bashat, D., Ben-Sira, L., Joskowicz, L., 2021. Fetal brain MRI measurements using a deep learning landmark network with reliability estimation, in: Sudre, C.H., et al. (Eds.), Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Perinatal Imaging, Placental and Preterm Image Analysis. UNSURE PIPPI 2021. Lect. Note...

  18. [18]

    MICCAI 2024 Challenge, 2024

    Fetal Tissue Annotation and Segmentation. MICCAI 2024 Challenge, 2024. https://fetachallenge.github.io (accessed 6 February 2026)

  19. [19]

    Advances in automated fetal brain MRI segmentation and biometry: insights from the FeTA 2024 challenge

    Zalevskyi, V., Sanchez, T., Kaandorp, M., Roulet, M., Fajardo‑Rojas, D., Liu, L., Hutter, J., Bran Li, H., Barkovich, M.J., Ji, H., Preložnik, D., Špiclin, Ž., et al., 2026. Advances in automated fetal brain MRI segmentation and biometry: insights from the FeTA 2024 challenge. Med. Image Anal. 109, 103941. https://doi.org/10.1016/j.media.2026.103941

  20. [20]

    https://www.developingconnectome.org (accessed 6 February 2026)

    The Developing Human Connectome Project. https://www.developingconnectome.org (accessed 6 February 2026)

  21. [21]

    An efficient total variation algorithm for super-resolution in fetal brain MRI with adaptive regularization

    Tourbier, S., Bresson, X., Hagmann, P., Thiran, J.-P., Meuli, R., Bach Cuadra, M., 2015. An efficient total variation algorithm for super-resolution in fetal brain MRI with adaptive regularization. Neuroimage 118, 584–97. https://doi.org/10.1016/j.neuroimage.2015.06.018

  22. [22]

    The Medical Image Analysis Laboratory Super -Resolution ToolKit (MIALSRTK) v2.0.0 (Version 2.0.0) [software]

    Tourbier S, De Dumast P, Kebiri H, 2020. The Medical Image Analysis Laboratory Super -Resolution ToolKit (MIALSRTK) v2.0.0 (Version 2.0.0) [software]. Zenodo. https://doi.org/10.5281/zenodo.7612119

  23. [23]

    Image Registration Toolkit (IRTK) (Version accessed 2025) [software]

    Rueckert, D., Sonoda, L.I., Hayes, C., Hill, D.L.G., Leach, M.O., Hawkes, D.J., 1999. Image Registration Toolkit (IRTK) (Version accessed 2025) [software]. https://biomedia.doc.ic.ac.uk/software/irtk/

  24. [24]

    https://www.synapse.org (accessed 6 February 2026)

    Synapse. https://www.synapse.org (accessed 6 February 2026)

  25. [25]

    The developing human connectome project (dHCP): fetal acquisition protocol

    Price, A.N., Cordero-Grande, L., Hughes, E., Hiscocks, S., Green, E., McCabe, L., et al., 2019. The developing human connectome project (dHCP): fetal acquisition protocol. Proc. Annu. Meet. Int. Soc. Magn. Reson. Med. 244

  26. [26]

    Automatic whole brain MRI segmentation of the developing neonatal brain

    Makropoulos, A., Gousias, I.S., Ledig, C., Aljabar, P., Serag, A., Hajnal, J.V., et al., 2014. Automatic whole brain MRI segmentation of the developing neonatal brain. IEEE Trans. Med. Imaging 33, 1818 –31. https://doi.org/10.1109/TMI.2014.2322280

  27. [27]

    Region Annotation With Expectation-Maximization (Draw-EM) (Version accessed 2025) [software]

    Makropoulos, A., Gousias, I.S., Ledig, C., Aljabar, P., Serag, A., Hajnal, J.V., et al., 2014. Region Annotation With Expectation-Maximization (Draw-EM) (Version accessed 2025) [software]. https://biomedia.doc.ic.ac.uk/software/draw - em/

  28. [28]

    BOUNTI: brain vOlumetry and aUtomated parcellatioN for 3D feTal MRI

    Uus, A., Kyriakopoulou, V., Makropoulos, A., Fukami-Gartner, A., Cromb, D., Davidson, A., et al., 2023. BOUNTI: brain vOlumetry and aUtomated parcellatioN for 3D feTal MRI. bioRxiv. https://doi.org/10.1101/2023.04.18.537347

  29. [29]

    U., Kyriakopoulou, V., Makropoulos, A., Fukami-Gartner, A., Cromb, D., Davidson, A., et al., 2023

    Uus, A. U., Kyriakopoulou, V., Makropoulos, A., Fukami-Gartner, A., Cromb, D., Davidson, A., et al., 2023. BOUNTI – Brain vOlumetry and aUtomated parcellatioN for 3D feTal MRI (Version accessed 2026) [Software]. Docker Hub. https://hub.docker.com/r/fetalsvrtk/segmentation

  30. [30]

    Fetal brain biometric measurements on 3D super-resolution reconstructed T2-weighted MRI: an intra- and inter-observer agreement study

    Khawam, M., de Dumast, P., Deman, P., Kebiri, H., Yu, T., Tourbier, S., et al., 2021. Fetal brain biometric measurements on 3D super-resolution reconstructed T2-weighted MRI: an intra- and inter-observer agreement study. Front. Pediatr. 9, 639746. https://doi.org/10.3389/fped.2021.639746

  31. [31]

    Fetal biometry annotation guidelines, 2024.https://unils- my.sharepoint.com/:b:/g/personal/meritxell_bachcuadra_unil_ch/ERBHllIq6lJFguOyxB299r0BktZpymzvVr7SvWAwvflQ - A?e=BSsolN (accessed 6 February 2026)

  32. [32]

    Multi -channel spatio-temporal MRI atlas of the normal fetal brain development from the developing Human Connectome Project

    Uus, A., Kyriakopoulou, V., Cordero Grande, L., Christiaens, D., Pietsch, M., Price, A., et al., 2023. Multi -channel spatio-temporal MRI atlas of the normal fetal brain development from the developing Human Connectome Project. G -Node. https://doi.org/10.12751/g-node.ysgsy1

  33. [33]

    Advanced normalization tools (ANTS)

    Avants, B., Tustison, N., Song, G., 2008. Advanced normalization tools (ANTS). Insight J. 1 –35. https://doi.org/10.54294/uvnhin

  34. [34]

    Advanced Normalization Tools (ANTs) v2.5.3 (Version 2.5.3)[software]

    Avants, B., Tustison, N., Song, G., 2008. Advanced Normalization Tools (ANTs) v2.5.3 (Version 2.5.3)[software]. GitHub: https://github.com/ANTsX/ANTs

  35. [35]

    Deep learning-based optimization of field geometry for total marrow irradiation delivered with volumetric modulated arc therapy

    Lambri, N., Longari, G., Loiacono, D., Brioso, R.C., Crespi, L., Galdieri, C., et al., 2024. Deep learning-based optimization of field geometry for total marrow irradiation delivered with volumetric modulated arc therapy. Med. Phys. 51, 4402–12. https://doi.org/10.1002/mp.17089

  36. [36]

    A guideline of selecting and reporting intraclass correlation coefficients for reliability research

    Koo, T.K., Li, M.Y., 2016. A guideline of selecting and reporting intraclass correlation coefficients for reliability research. J. Chiropr. Med. 15, 155–63. https://doi.org/10.1016/j.jcm.2016.02.012

  37. [37]

    auto -proc-svrtk (Version accessed 2025) [software]

    Luis, A., Uus, A., Matthew, J., Arulkumaran, S., Egloff Collado, A., Kyriakopoulou, V., et al., 2025. auto -proc-svrtk (Version accessed 2025) [software]. GitHub: https://github.com/SVRTK/auto -proc-svrtk

  38. [38]

    Using Cliff’s delta as a non-parametric effect size measure: an accessible Web App and R tutorial

    Meissel, K., Yao, E.S., 2024. Using Cliff’s delta as a non-parametric effect size measure: an accessible Web App and R tutorial. Pract. Assess. Res. Eval. 29, 2. https://doi.org/10.7275/pare.1977

  39. [39]

    ISUOG practice guidelines: ultrasound assessment of fetal biometry and growth

    Salomon, L.J., Alfirevic, Z., Da Silva Costa, F., Deter, R.L., Figueras, F., Ghi, T., et al., 2019. ISUOG practice guidelines: ultrasound assessment of fetal biometry and growth. Ultrasound Obstet. Gynecol. 53, 715 –23. https://doi.org/10.1002/uog.20272

  40. [40]

    Fetal brain MRI: neurometrics, typical diagnoses, and resolving common dilemmas

    Brady, D., Schlatterer, S.D., Whitehead, M.T., 2022. Fetal brain MRI: neurometrics, typical diagnoses, and resolving common dilemmas. Br. J. Radiol. 96, 20211019. https://doi.org/10.1259/bjr.20211019

  41. [41]

    ISUOG practice guidelines (updated): performance of fetal magnetic resonance imaging

    Prayer, D., Malinger, G., De Catte, L., De Keersmaecker, B., Gonçalves, L.F., Kasprian, G., et al., 2023. ISUOG practice guidelines (updated): performance of fetal magnetic resonance imaging. Ultrasound Obstet. Gynecol. 61, 278 –87. https://doi.org/10.1002/uog.26129

  42. [42]

    The reliability of fetal MRI in the assessment of brain malformations

    Frick, N., Meuli, R., O’Gorman Tuura, R., Muehlethaler, V., Wiest, R., Huisman, T.A.G.M., 2015. The reliability of fetal MRI in the assessment of brain malformations. Fetal Diagn. Ther. 37, 93–101. https://doi.org/10.1159/000363652

  43. [43]

    Prenatal magnetic resonance imaging: brain normal linear biometric values below 24 gestational weeks

    Parazzini, C., Righini, A., Rustico, M., Consonni, D., Triulzi, F., 2008. Prenatal magnetic resonance imaging: brain normal linear biometric values below 24 gestational weeks. Neuroradiology 50, 877 –83. https://doi.org/10.1007/s00234- 008-0421-7

  44. [44]

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

    Zhang, J., Liu, M., Shen, D., 2017. Detecting anatomical landmarks from limited medical imaging data using two -stage task-oriented deep neural networks. IEEE Trans. Pattern Anal. Mach. Intell. 26, 4753 –64

  45. [45]

    Uncertainty estimation in landmark localization based on Gaussian heatmaps

    Payer, C., Urschler, M., Bischof, H., Štern, D., 2020. Uncertainty estimation in landmark localization based on Gaussian heatmaps. In: Sudre, C.H., et al. (Eds.), Uncertainty for Safe Utilization of Machine Learning in Medical Imaging. Lecture Notes in Computer Science, Springer, Cham

  46. [46]

    BiometryNet: Landmark-based Fetal Biometry Estimation from Standard Ultrasound Planes

    Avisdris, N., Joskowicz, L., Dromey, B., David, A.L., Peebles, D.M., Stoyanov, D., et al., 2022. BiometryNet: landmark-based fetal biometry estimation from standard ultrasound planes. arXiv:2206.14678. https://doi.org/10.48550/arXiv.2206.14678

  47. [47]

    Cerebral biometry in fetal magnetic resonance imaging: new reference data

    Tilea, B., Alberti, C., Adamsbaum, C., Armoogum, P., Oury, J.F., Cabrol, D., et al., 2009. Cerebral biometry in fetal magnetic resonance imaging: new reference data. Ultrasound Obstet. Gynecol. 33, 173 –81. https://doi.org/10.1002/uog.6276. Supplementary Table S1 Table S1. Mapping of the 19 tissue labels produced by BOUNTI to the 8-tissue label maps used ...

  48. [2355]

    https://doi.org/10.3390/diagnostics13142355

This paper was first reviewed by deepseek-v4-flash on August 5, 2026.