pith. machine review for the scientific record. sign in

arxiv: 2605.09575 · v1 · submitted 2026-05-10 · 📡 eess.IV · cs.CV

Recognition: no theorem link

Annotation-free deep learning for detection and segmentation of fetal germinal matrix-intraventricular hemorrhage in brain MRI

Gang Ning, Haibo Qu, Haoxiang Li, Hongjia Yang, Hua Lai, Jialan Zheng, Juncheng Zhu, Kasidit Anmahapong, Mingxuan Liu, Min Kang, Nan Sun, Qiyuan Tian, Rong Hu, Xiaoling Zhou, Yan Song, Yifei Chen, Yijin Li, Yi Liao, Yingqi Hao, Zihan Li

Authors on Pith no claims yet

Pith reviewed 2026-05-12 04:46 UTC · model grok-4.3

classification 📡 eess.IV cs.CV
keywords fetal MRIGMH-IVHhemorrhage detectionannotation-free learningimage synthesisbrain segmentationdeep learningprenatal diagnosis
0
0 comments X

The pith

An annotation-free deep learning model detects and segments fetal brain hemorrhages on MRI by training only on synthesized pseudo-images.

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

The paper introduces FreeHemoSeg to automate detection and segmentation of prenatal GMH-IVH, a major cause of infant harm that currently requires labor-intensive manual review of fetal brain MRIs. Standard deep learning approaches demand large sets of annotated real cases, which are difficult to assemble for this condition. Instead the authors generate pseudo hemorrhage images from normal fetal scans by applying medical knowledge of typical hemorrhage locations and appearances, then train the model exclusively on those images. The resulting system achieves high sensitivity and specificity on both internal and external real-world cases while also boosting radiologist accuracy and cutting review time.

Core claim

FreeHemoSeg is an annotation-free deep learning framework trained solely on pseudo GMH-IVH images synthesized from normal fetal T2-weighted MRI data according to medical priors; on held-out real cases it reaches internal sensitivity 0.914 and specificity 0.966 with DSC 0.559, external sensitivity 0.824 and specificity 0.943 with DSC 0.512, outperforming both supervised and unsupervised baselines and improving radiologist performance when used as assistance.

What carries the argument

Synthesis of pseudo GMH-IVH images from normal fetal MRI scans guided by medical priors on hemorrhage site and signal characteristics, which then serve as the sole training data for the FreeHemoSeg detection and segmentation network.

If this is right

  • Automated detection and segmentation of fetal brain hemorrhages becomes feasible without any manual annotations.
  • Radiologists achieve higher sensitivity and greater diagnostic confidence while spending less time per case.
  • Performance remains strong across data collected at multiple hospitals.
  • Earlier and more consistent identification of GMH-IVH can support timely clinical decisions.

Where Pith is reading between the lines

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

  • Medical priors can substitute for scarce annotations when building AI systems for rare fetal imaging findings.
  • The synthesis strategy may extend to other low-prevalence prenatal or pediatric brain conditions.
  • Lowering the annotation barrier could speed development of diagnostic tools for uncommon medical events.

Load-bearing premise

Pseudo hemorrhage images created from normal scans using medical priors resemble real GMH-IVH closely enough that a model trained on them will generalize to actual clinical cases.

What would settle it

FreeHemoSeg is evaluated on a large independently annotated collection of real GMH-IVH cases and shows substantially lower sensitivity or specificity than the reported validation numbers.

Figures

Figures reproduced from arXiv: 2605.09575 by Gang Ning, Haibo Qu, Haoxiang Li, Hongjia Yang, Hua Lai, Jialan Zheng, Juncheng Zhu, Kasidit Anmahapong, Mingxuan Liu, Min Kang, Nan Sun, Qiyuan Tian, Rong Hu, Xiaoling Zhou, Yan Song, Yifei Chen, Yijin Li, Yi Liao, Yingqi Hao, Zihan Li.

Figure 2
Figure 2. Figure 2: FreeHemoSeg Framework. (A) Overview of the three-stage pipeline. (B) Stage 1 – Data Synthesis: Pseudo GMH‑IVH slices are synthesized from normal fetal brain images. (C) Stage 2 – Model Training: The synthesized slices are used to train segmentation models and fine‑tune the SAM model. (D) Stage 3 – Model Inference: Segmentation probability heatmaps are generated for diagnosis and initial segmentation, and t… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison between Empirical and Pseudo GMH-IVH Slices. Enlarged regions highlight hemorrhagic areas of empirical GMH-IVH slices (A) and pseudo GMH-IVH slices (B). Slice-level comparison of hemorrhagic region sizes (C) and image intensity values (D) within the hemorrhage regions [PITH_FULL_IMAGE:figures/full_fig_p025_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Diagnosis Performance Analysis. (A-F) Internal validation; (G-L) External validation. (A, G) Case-level receiver operating characteristic (ROC) curves for germinal matrix-intraventricular hemorrhage (GMH-IVH) diagnosis. (B, H) Case-level precision-recall (PR) curves for GMH-IVH diagnosis. (C, I) Confusion matrices showing case-level classification results for FreeHemoSeg at the threshold determined by maxi… view at source ↗
Figure 1
Figure 1. Figure 1: Flow Diagram of Training and Validation Datasets. Abbreviations: SL Model, supervised learning model; UAD, unsupervised anomaly detection; GMH-IVH, germinal matrix-intraventricular hemorrhage [PITH_FULL_IMAGE:figures/full_fig_p027_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FreeHemoSeg Framework. (A) Overview of the three-stage pipeline. (B) Stage 1 – Data Synthesis: Pseudo GMH‑IVH slices are synthesized from normal fetal brain images. (C) Stage 2 – Model Training: The synthesized slices are used to train segmentation models and fine‑tune the SAM model. (D) Stage 3 – Model Inference: Segmentation probability heatmaps are generated for diagnosis and initial segmentation, and t… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison between Empirical and Pseudo GMH-IVH Slices. Enlarged regions highlight hemorrhagic areas of empirical GMH-IVH slices (A) and pseudo GMH-IVH slices (B). Slice-level comparison of hemorrhagic region sizes (C) and image intensity values (D) within the hemorrhage regions [PITH_FULL_IMAGE:figures/full_fig_p029_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Diagnosis Performance Analysis. (A-F) Internal validation; (G-L) External validation. (A, G) Case-level receiver operating characteristic (ROC) curves for germinal matrix-intraventricular hemorrhage (GMH-IVH) diagnosis. (B, H) Case-level precision-recall (PR) curves for GMH-IVH diagnosis. (C, I) Confusion matrices showing case-level classification results for FreeHemoSeg at the threshold determined by maxi… view at source ↗
read the original abstract

Background: Prenatal germinal matrix-intraventricular hemorrhage (GMH-IVH) is a leading cause of infant mortality and neurodevelopmental impairment. Manual diagnosis and lesion segmentation are labor-intensive and error-prone. Deep learning models offer potential for automation but typically require large annotated datasets, which are challenging to obtain. Purpose: To develop and validate an annotation-free deep learning framework for automated detection and segmentation of GMH-IVH on brain MRI. Materials and Methods: This retrospective study analyzed 2D T2-weighted MRI data from pregnant women collected from October 2015 to October 2023 at one hospital (internal validation) and two hospitals (external validation). Eligible participants included healthy fetuses and those with GMH-IVH. FreeHemoSeg was developed and trained using pseudo GMH-IVH images synthesized from normal fetal data guided by medical priors. Primary outcomes included diagnostic accuracy (area under the ROC curve [AUROC], sensitivity, specificity) and segmentation accuracy (Dice similarity coefficient [DSC]). A reader study evaluated clinical utility. Results: A total of 1674 stacks from 558 pregnant women were analyzed. FreeHemoSeg achieved the highest performance in both internal (sensitivity: 0.914, 95% CI 0.869-0.945; specificity: 0.966, 95% CI 0.946-0.978; DSC: 0.559, 95% CI 0.546-0.571) and external validation (sensitivity: 0.824, 95% CI 0.739-0.885; specificity: 0.943, 95% CI 0.913-0.964; DSC: 0.512, 95% CI 0.497-0.526), outperforming supervised and unsupervised methods. FreeHemoSeg assistance improved radiologists' sensitivity (from 0.882 to 0.941-1.000) and diagnostic confidence while reducing interpretation time by 16.0-52.7%. Conclusion: FreeHemoSeg accurately detects and localizes fetal brain hemorrhages without annotated training data, enabling earlier diagnosis and supporting timely clinical management.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript presents FreeHemoSeg, an annotation-free deep learning framework for automated detection and segmentation of fetal GMH-IVH on T2-weighted brain MRI. The model is trained exclusively on pseudo-GMH-IVH images synthesized from normal fetal scans using medical priors, then evaluated on internal (one hospital) and external (two hospitals) retrospective datasets of 1674 stacks from 558 pregnant women. It reports strong diagnostic performance (internal sensitivity 0.914/specificity 0.966; external 0.824/0.943) with moderate segmentation (DSC 0.559 internal, 0.512 external), outperforming supervised and unsupervised baselines, and demonstrates improved radiologist sensitivity and reduced interpretation time in a reader study.

Significance. If the synthesized training data generalizes to real lesions, this annotation-free approach would be highly significant for prenatal imaging, where manual annotations for rare conditions like GMH-IVH are scarce and labor-intensive. The multi-site external validation, outperformance over baselines, and reader-study gains (sensitivity increase to 0.941-1.000 and 16-52.7% time reduction) provide concrete evidence of potential clinical utility for earlier diagnosis. The work's strength lies in its reproducible synthesis pipeline and falsifiable multi-center metrics, though moderate DSC values limit claims of precise localization.

major comments (1)
  1. [Methods (pseudo GMH-IVH synthesis)] Methods section on pseudo GMH-IVH image synthesis: The annotation-free claim is load-bearing on the assumption that synthesized images capture real GMH-IVH intensity profiles, textures, shapes, locations, and MRI artifacts across gestational ages and stages. No quantitative fidelity validation (intensity histograms, shape descriptors, or perceptual metrics) is reported. The external performance drop (sensitivity 0.914 to 0.824, DSC 0.559 to 0.512) and only moderate DSC indicate possible exploitation of synthesis-specific cues rather than true lesion features, leaving the domain-gap assumption untested and requiring empirical support such as an ablation study.
minor comments (1)
  1. [Abstract] Abstract and Results: The abstract states outperformance over baselines but does not quantify the margins or name the exact supervised/unsupervised methods; adding these details would improve clarity without altering the central claims.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the major comment on the pseudo-GMH-IVH synthesis validation below and will incorporate the suggested empirical support in the revised version.

read point-by-point responses

  1. Referee: Methods section on pseudo GMH-IVH image synthesis: The annotation-free claim is load-bearing on the assumption that synthesized images capture real GMH-IVH intensity profiles, textures, shapes, locations, and MRI artifacts across gestational ages and stages. No quantitative fidelity validation (intensity histograms, shape descriptors, or perceptual metrics) is reported. The external performance drop (sensitivity 0.914 to 0.824, DSC 0.559 to 0.512) and only moderate DSC indicate possible exploitation of synthesis-specific cues rather than true lesion features, leaving the domain-gap assumption untested and requiring empirical support such as an ablation study.

    Authors: We agree that quantitative fidelity metrics would strengthen the evidence that our synthesis captures real lesion characteristics. In the revision, we will add comparisons of intensity histograms, Haralick texture features, and shape descriptors (e.g., area, eccentricity) between synthesized and real GMH-IVH lesions drawn from the annotated validation cases. The observed external performance drop is expected given inter-site differences in scanner vendors, field strengths, and acquisition protocols; importantly, FreeHemoSeg still outperforms all supervised and unsupervised baselines on the external set, which would be unlikely if the model were merely exploiting synthesis-specific artifacts. We will further include an ablation study that systematically removes individual synthesis priors (intensity perturbation, location sampling, gestational-age scaling) and re-evaluates external generalization to isolate their contribution. These additions will directly test the domain-gap assumption. revision: yes

Circularity Check

0 steps flagged

No significant circularity in annotation-free training and validation pipeline

full rationale

The paper trains FreeHemoSeg exclusively on pseudo GMH-IVH images synthesized from normal fetal T2-weighted MRI using medical priors, then reports performance on independent real multi-site clinical cases (internal and external validation cohorts). No equations, derivations, or performance metrics reduce by construction to the synthesis inputs; the external-validation results (AUROC, sensitivity, DSC) are measured on held-out real data and thus constitute independent evidence. No load-bearing self-citations, uniqueness theorems, or ansatzes imported from prior author work are invoked to justify the central claim. The separation between synthetic training data and real evaluation data keeps the annotation-free assertion self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the assumption that medical priors produce sufficiently realistic pseudo-hemorrhage images for training; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Medical priors can guide synthesis of realistic pseudo GMH-IVH images from normal fetal MRI data
    This assumption underpins the entire annotation-free training strategy.

pith-pipeline@v0.9.0 · 5791 in / 1246 out tokens · 33587 ms · 2026-05-12T04:46:38.297089+00:00 · methodology

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

44 extracted references · 44 canonical work pages

  1. [1]

    Fetal intracranial hemorrhage: role of fetal MRI: Fetal intracranial hemorrhage and fetal MRI

    Sanapo L, Whitehead MT, Bulas DI, et al. Fetal intracranial hemorrhage: role of fetal MRI: Fetal intracranial hemorrhage and fetal MRI. Prenat Diagn. 2017;37(8):827–836. doi: 10.1002/pd.5096

    work page doi:10.1002/pd.5096 2017

  2. [2]

    Early and Late Complications of Germinal Matrix-Intraventricular Haemorrhage in the Preterm Infant: What Is New? Neonatology

    Brouwer AJ, Groenendaal F, Benders MJNL, De Vries LS. Early and Late Complications of Germinal Matrix-Intraventricular Haemorrhage in the Preterm Infant: What Is New? Neonatology. 2014;106(4):296–303. doi: 10.1159/000365127

    work page doi:10.1159/000365127 2014

  3. [3]

    Fetal intraventricular hemorrhage and periventricular hemorrhagic venous infarction: time for dedicated classification system

    Hadi E, Haddad L, Levy M, et al. Fetal intraventricular hemorrhage and periventricular hemorrhagic venous infarction: time for dedicated classification system. Ultrasound Obstet Gynecol Off J Int Soc Ultrasound Obstet Gynecol. England; 2024;64(3):285–

    work page 2024

  4. [4]

    doi: 10.1002/uog.27613

    work page doi:10.1002/uog.27613

  5. [5]

    Fetal intracranial hemorrhage (fetal stroke): does grade matter? Ultrasound Obstet Gynecol

    Elchalal U, Yagel S, Gomori JM, et al. Fetal intracranial hemorrhage (fetal stroke): does grade matter? Ultrasound Obstet Gynecol. 2005;26(3):233–243. doi: 10.1002/uog.1969

    work page doi:10.1002/uog.1969 2005

  6. [6]

    Antenatal diagnosis of fetal intraventricular hemorrhage: systematic review and meta‐analysis

    Dunbar MJ, Woodward K, Leijser LM, Kirton A. Antenatal diagnosis of fetal intraventricular hemorrhage: systematic review and meta‐analysis. Dev Med Child Neurol. 2021;63(2):144–155. doi: 10.1111/dmcn.14713

    work page doi:10.1111/dmcn.14713 2021

  7. [7]

    Prenatal diagnosis and postnatal outcome of fetal intracranial hemorrhage: a single-center experience

    Kim S, Jung YJ, Baik J, et al. Prenatal diagnosis and postnatal outcome of fetal intracranial hemorrhage: a single-center experience. Obstet Gynecol Sci. 2024;67(4):393–403. doi: 10.5468/ogs.24097

    work page doi:10.5468/ogs.24097 2024

  8. [8]

    Epstein, Kline-Fath BM, Zhang B, et al

    K.N. Epstein, Kline-Fath BM, Zhang B, et al. Prenatal Evaluation of Intracranial Hemorrhage on Fetal MRI: A Retrospective Review. Am J Neuroradiol. 2021;42(12):2222–2228. doi: 10.3174/ajnr.A7320

    work page doi:10.3174/ajnr.a7320 2021

  9. [9]

    Fetal intracranial hemorrhage and infarct: Main sonographic and MRI characteristics: A review article

    Moradi B, Ardestani RM, Shirazi M, Eslamian L, Kazemi MA. Fetal intracranial hemorrhage and infarct: Main sonographic and MRI characteristics: A review article. Eur J Obstet Gynecol Reprod Biol X. 2024;24:100351. doi: 10.1016/j.eurox.2024.100351

    work page doi:10.1016/j.eurox.2024.100351 2024

  10. [10]

    SmaRT: Style-Modulated Robust Test-Time Adaptation for Cross-Domain Brain Tumor Segmentation in MRI

    Wang Y, Chen Y, Jiang S, et al. SmaRT: Style-Modulated Robust Test-Time Adaptation for Cross-Domain Brain Tumor Segmentation in MRI. ArXiv E-Prints. 2025;arXiv:2509.17925. doi: 10.48550/arXiv.2509.17925

    work page doi:10.48550/arxiv.2509.17925 2025

  11. [11]

    DR-TTA: Dynamic and Robust Test-Time Adaptation Under Low-Quality Mri Conditions for Brain Tumor Segmentation

    Wang Y, Chen Y, Jiang S, et al. DR-TTA: Dynamic and Robust Test-Time Adaptation Under Low-Quality Mri Conditions for Brain Tumor Segmentation. 2025 IEEE Int Conf Bioinforma Biomed BIBM. 2025. p. 2899–2906. doi: 10.1109/BIBM66473.2025.11356381

    work page doi:10.1109/bibm66473.2025.11356381 2025

  12. [12]

    Vggsound: A Large-Scale Audio-Visual Dataset

    Huang H, Lin L, Tong R, et al. UNet 3+: A Full-Scale Connected UNet for Medical Image Segmentation. ICASSP 2020 - 2020 IEEE Int Conf Acoust Speech Signal Process ICASSP. Barcelona, Spain: IEEE; 2020. p. 1055–1059. doi: 10.1109/ICASSP40776.2020.9053405

    work page doi:10.1109/icassp40776.2020.9053405 2020

  13. [13]

    H-denseunet: Hybrid densely connected unet for liver and tumor segmentation from ct volumes

    Li X, Chen H, Qi X, Dou Q, Fu C-W, Heng P-A. H-DenseUNet: Hybrid Densely Connected UNet for Liver and Tumor Segmentation From CT Volumes. IEEE Trans Med Imaging. 2018;37(12):2663–2674. doi: 10.1109/TMI.2018.2845918

    work page doi:10.1109/tmi.2018.2845918 2018

  14. [14]

    Low-grade intraventricular hemorrhage: is 16 ultrasound good enough? J Matern Fetal Neonatal Med

    Alessandro Parodi AR Giovanni Morana, Maria S Severino, Mariya Malova, Anna R Natalizia, Andrea Sannia, Ramenghi LA. Low-grade intraventricular hemorrhage: is 16 ultrasound good enough? J Matern Fetal Neonatal Med. Taylor & Francis; 2015;28(sup1):2261–2264. doi: 10.3109/14767058.2013.796162

    work page doi:10.3109/14767058.2013.796162 2015

  15. [15]

    A Labeling-Free Approach to Supervising Deep Neural Networks for Retinal Blood Vessel Segmentation

    Chen Y. A Labeling-Free Approach to Supervising Deep Neural Networks for Retinal Blood Vessel Segmentation. arXiv; 2017. http://arxiv.org/abs/1704.07502. Accessed August 19, 2024

    work page arXiv 2017

  16. [16]

    Proxy-Bridged Image Reconstruction Network for Anomaly Detection in Medical Images

    Zhou K, Li J, Luo W, et al. Proxy-Bridged Image Reconstruction Network for Anomaly Detection in Medical Images. IEEE Trans Med Imaging. 2022;41(3):582–594. doi: 10.1109/TMI.2021.3118223

    work page doi:10.1109/tmi.2021.3118223 2022

  17. [18]

    Synthetic Data for Robust Stroke Segmentation

    Chalcroft L, Pappas I, Price CJ, Ashburner J. Synthetic Data for Robust Stroke Segmentation. arXiv; 2024. http://arxiv.org/abs/2404.01946. Accessed August 19, 2024

    work page arXiv 2024

  18. [20]

    Label-Free Segmentation of COVID-19 Lesions in Lung CT

    Yao Q, Xiao L, Liu P, Zhou SK. Label-Free Segmentation of COVID-19 Lesions in Lung CT. IEEE Trans Med Imaging. 2021;40(10):2808–2819. doi: 10.1109/TMI.2021.3066161

    work page doi:10.1109/tmi.2021.3066161 2021

  19. [21]

    LeFusion: Synthesizing Myocardial Pathology on Cardiac MRI via Lesion-Focus Diffusion Models

    Zhang H, Yang J, Wan S, Fua P. LeFusion: Synthesizing Myocardial Pathology on Cardiac MRI via Lesion-Focus Diffusion Models. arXiv; 2024. http://arxiv.org/abs/2403.14066. Accessed August 19, 2024

    work page arXiv 2024

  20. [22]

    Local Style Transfer via Latent Space Manipulation for Cross-Disease Lesion Segmentation

    Lyu F, Ye M, Yip TC-F, Wong GL-H, Yuen PC. Local Style Transfer via Latent Space Manipulation for Cross-Disease Lesion Segmentation. IEEE J Biomed Health Inform. 2024;28(1):273–284. doi: 10.1109/JBHI.2023.3327726

    work page doi:10.1109/jbhi.2023.3327726 2024

  21. [27]

    Beyond Adapting SAM: Towards End-to-End Ultrasound Image Segmentation via Auto Prompting

    Lin X, Xiang Y, Yu L, Yan Z. Beyond Adapting SAM: Towards End-to-End Ultrasound Image Segmentation via Auto Prompting. In: Linguraru MG, Dou Q, Feragen A, et al., editors. Med Image Comput Comput Assist Interv – MICCAI 2024. Cham: Springer Nature Switzerland; 2024. p. 24–34. 17

    work page 2024

  22. [31]

    Transformer-based Deep Neural Network for Breast Cancer Classification on Digital Breast Tomosynthesis Images

    Lee W, Lee H, Lee H, Park EK, Nam H, Kooi T. Transformer-based Deep Neural Network for Breast Cancer Classification on Digital Breast Tomosynthesis Images. Radiol Artif Intell. 2023;5(3):e220159. doi: 10.1148/ryai.220159

    work page doi:10.1148/ryai.220159 2023

  23. [32]

    Fetal Germinal Matrix and Intraventricular Hemorrhage

    Morioka T, Hashiguchi K, Nagata S, et al. Fetal Germinal Matrix and Intraventricular Hemorrhage. Pediatr Neurosurg. 2006;42(6):354–361. doi: 10.1159/000095565

    work page doi:10.1159/000095565 2006

  24. [33]

    FETAL INTRAVENTRICULAR HAEMORRHAGE: DIFFERENT APPROACHES NEEDED? INSIGHTS ON GASLINI’S EXPERIENCE

    Ramenghi LA, Samuele C. FETAL INTRAVENTRICULAR HAEMORRHAGE: DIFFERENT APPROACHES NEEDED? INSIGHTS ON GASLINI’S EXPERIENCE

    work page

  25. [34]

    Gonçalves LF, Lee W, Mody S, Shetty A, Sangi-Haghpeykar H, Romero R. Diagnostic accuracy of ultrasonography and magnetic resonance imaging for the detection of fetal anomalies: a blinded case-control study: Accuracy of US and MRI in detecting fetal anomalies. Ultrasound Obstet Gynecol. 2016;48(2):185–192. doi: 10.1002/uog.15774

    work page doi:10.1002/uog.15774 2016

  26. [35]

    Real-time automatic fetal brain extraction in fetal MRI by deep learning

    Salehi SSM, Hashemi SR, Velasco-Annis C, et al. Real-time automatic fetal brain extraction in fetal MRI by deep learning. 2018 IEEE 15th Int Symp Biomed Imaging ISBI 2018. 2018. p. 720–724. doi: 10.1109/ISBI.2018.8363675

    work page doi:10.1109/isbi.2018.8363675 2018

  27. [36]

    reverse distillation

    Prayer D, Malinger G, De Catte L, et al. ISUOG Practice Guidelines (updated): performance of fetal magnetic resonance imaging. Ultrasound Obstet Gynecol. 2023;61(2):278–287. doi: 10.1002/uog.26129. 18 Table 1. Baseline Characteristics of the Study Population Characteristic Training Dataset A Training Dataset B Validation Dataset A Validation Dataset B Fun...

    work page doi:10.1002/uog.26129 2023

  28. [37]

    An automated framework for localization, segmentation and super-resolution reconstruction of fetal brain MRI

    Ebner M, Wang G, Li W, et al. An automated framework for localization, segmentation and super-resolution reconstruction of fetal brain MRI. NeuroImage. 2020;206:116324. doi: 10.1016/j.neuroimage.2019.116324

    work page doi:10.1016/j.neuroimage.2019.116324 2020

  29. [38]

    NeSVoR: Implicit Neural Representation for Slice-to- Volume Reconstruction in MRI

    Xu J, Moyer D, Gagoski B, et al. NeSVoR: Implicit Neural Representation for Slice-to- Volume Reconstruction in MRI. IEEE Trans Med Imaging. 2023;42(6):1707–1719. doi: 10.1109/TMI.2023.3236216

    work page doi:10.1109/tmi.2023.3236216 2023

  30. [40]

    A Dempster-Shafer approach to trustworthy AI with application to fetal brain MRI segmentation

    Fidon L, Aertsen M, Kofler F, et al. A Dempster-Shafer approach to trustworthy AI with application to fetal brain MRI segmentation. IEEE Trans Pattern Anal Mach Intell. 2024;1–12. doi: 10.1109/TPAMI.2023.3346330

    work page doi:10.1109/tpami.2023.3346330 2024

  31. [41]

    Incidence and evolution of subependymal and intraventricular hemorrhage: a study of infants with birth weights less than 1,500 gm

    Papile L-A, Burstein J, Burstein R, Koffler H. Incidence and evolution of subependymal and intraventricular hemorrhage: a study of infants with birth weights less than 1,500 gm. J Pediatr. Elsevier; 1978;92(4):529–534

    work page 1978

  32. [42]

    https://doi.org/10.1007/978-3- 642-13875-1

    Zhang Z, Deng H, Li X. Unsupervised Liver Tumor Segmentation with Pseudo Anomaly Synthesis. In: Wolterink JM, Svoboda D, Zhao C, Fernandez V, editors. Simul Synth Med Imaging. Cham: Springer Nature Switzerland; 2023. p. 86–96. doi: 10.1007/978-3- 031-44689-4_9

    work page doi:10.1007/978-3- 2023

  33. [43]

    VM-UNet: Vision Mamba UNet for Medical Image Segmentation

    Ruan J, Xiang S. VM-UNet: Vision Mamba UNet for Medical Image Segmentation. arXiv; 2024. http://arxiv.org/abs/2402.02491. Accessed August 19, 2024

    work page arXiv 2024

  34. [44]

    https:// arxiv.org/abs/2401.10166

    Liu Y, Tian Y, Zhao Y, et al. VMamba: Visual State Space Model. ArXiv. 2024;abs/2401.10166. https://api.semanticscholar.org/CorpusID:267035250

    work page arXiv 2024

  35. [45]

    Decoupled weight decay regularization

    Loshchilov I, Hutter F. Decoupled weight decay regularization. ArXiv Prepr ArXiv171105101. 2017

    work page 2017

  36. [46]

    Segment anything

    Kirillov A, Mintun E, Ravi N, et al. Segment anything. Proc IEEECVF Int Conf Comput Vis. 2023. p. 4015–4026

    work page 2023

  37. [47]

    Beyond Adapting SAM: Towards End-to-End Ultrasound Image Segmentation via Auto Prompting

    Lin X, Xiang Y, Yu L, Yan Z. Beyond Adapting SAM: Towards End-to-End Ultrasound Image Segmentation via Auto Prompting. In: Linguraru MG, Dou Q, Feragen A, et al., editors. Med Image Comput Comput Assist Interv – MICCAI 2024. Cham: Springer Nature Switzerland; 2024. p. 24–34

    work page 2024

  38. [48]

    Adam: A method for stochastic optimization

    Kingma DP. Adam: A method for stochastic optimization. ArXiv Prepr ArXiv14126980. 2014

    work page 2014

  39. [49]

    Bias in Unsupervised Anomaly Detection in Brain MRI

    Bercea CI, Puyol-Antón E, Wiestler B, Rueckert D, Schnabel JA, King AP. Bias in Unsupervised Anomaly Detection in Brain MRI. arXiv; 2023. http://arxiv.org/abs/2308.13861. Accessed August 29, 2023. 25

    work page arXiv 2023

  40. [50]

    Anomaly Detection for Medical Images Using Teacher- Student Model with Skip Connections and Multi-scale Anomaly Consistency

    Liu M, Jiao Y, Lu J, Chen H. Anomaly Detection for Medical Images Using Teacher- Student Model with Skip Connections and Multi-scale Anomaly Consistency. IEEE Trans Instrum Meas. 2024;1–1. doi: 10.1109/TIM.2024.3406792

    work page doi:10.1109/tim.2024.3406792 2024

  41. [51]

    Informative knowledge distillation for image anomaly segmentation

    Cao Y, Wan Q, Shen W, Gao L. Informative knowledge distillation for image anomaly segmentation. Knowl-Based Syst. 2022;248:108846. doi: 10.1016/j.knosys.2022.108846

    work page doi:10.1016/j.knosys.2022.108846 2022

  42. [52]

    Shadows can be

    Deng H, Li X. Anomaly Detection via Reverse Distillation from One-Class Embedding. 2022 IEEECVF Conf Comput Vis Pattern Recognit CVPR. New Orleans, LA, USA: IEEE; 2022. p. 9727–9736. doi: 10.1109/CVPR52688.2022.00951

    work page doi:10.1109/cvpr52688.2022.00951 2022

  43. [53]

    Deep high-resolution representation learning for visual recognition

    Wang J, Sun K, Cheng T, et al. Deep high-resolution representation learning for visual recognition. IEEE Trans Pattern Anal Mach Intell. IEEE; 2020;43(10):3349–3364

    work page 2020

  44. [54]

    Wide Residual Networks

    Zagoruyko S, Komodakis N. Wide Residual Networks. ArXiv Prepr ArXiv160507146. 2016

    work page 2016