CoChat
Sign Up
Cited by 13Medical Image Segmentation TechniquesArtificial Intelligence in Healthcare and EducationMedical Imaging and Analysis

Minimally interactive placenta segmentation from three-dimensional ultrasound images

Open access
Oguz, Ipek, Yushkevich, Natalie, Pouch, Alison M., Oguz, Baris U., Wang, Jiancong, Parameshwaran, Shobhana, Gee, James C., Yushkevich, Paul A. +1 more·2020·Journal of Medical Imaging
DOI 10.1117/1.jmi.7.1.014004PMC PMC7035878
Open this paper in CoChat — free account

Abstract

Purpose: Placental size in early pregnancy has been associated with important clinical outcomes, including fetal growth. However, extraction of placental size from three-dimensional ultrasound (3DUS) requires time-consuming interactive segmentation methods and is prone to user variability. We propose a semiautomated segmentation technique that requires minimal user input to robustly measure placental volume from 3DUS images. Approach: For semiautomated segmentation, a single, central 2D slice was manually annotated to initialize an automated multi-atlas label fusion (MALF) algorithm. The dataset consisted of 47 3DUS volumes obtained at 11 to 14 weeks in singleton pregnancies (28 anterior and 19 posterior). Twenty-six of these subjects were imaged twice within the same session. Dice overlap and surface distance were used to quantify the automated segmentation accuracy compared to expert manual segmentations. The mean placental volume measurements obtained by our method and VOCAL (virtual organ computer-aided analysis), a leading commercial semiautomated method, were compared to the manual reference set. The test–retest reliability was also assessed. Results: The overlap between our automated segmentation and manual (mean Dice: 0.824 ± 0.061, median: 0.831) was within the range reported by other methods requiring extensive manual input. The average surface distance was 1.66 ± 0.96 mm. The correlation coefficient between test–retest volumes was r = 0.88, and the intraclass correlation was ICC ( 1 ) = 0.86. Conclusions: MALF is a promising method that can allow accurate and reliable segmentation of the placenta with minimal user interaction. Further refinement of this technique may allow for placental biometry to be incorporated into clinical pregnancy surveillance.

AI snapshot

Written from the paper’s full text.

TL;DR

A minimally interactive multi-atlas label fusion method segments the placenta from first-trimester 3D ultrasound using only a single manually annotated 2D slice as input. It achieved Dice overlap of 0.824 ± 0.061 with expert manual segmentations and test–retest reliability of r = 0.88 / ICC(1) = 0.86.

Key findings

  • Overall Dice overlap between automated and manual segmentations was 0.824 ± 0.061 (median 0.831); anterior placentas: 0.838 ± 0.063 (median 0.852), posterior placentas: 0.803 ± 0.055 (median 0.821).
  • Mean surface distance was 1.66 ± 0.96 mm overall, 1.61 ± 1.17 mm for anterior and 1.73 ± 0.53 mm for posterior placentas.
  • Mean placental volume from the proposed method was 79.0 ± 18.2 cc, not significantly different from manual reference (80.7 ± 30.5 cc; p > 0.54), whereas VOCAL gave 69.7 ± 29.7 cc, significantly smaller than manual (p < 1 × 10⁻⁷).
  • Test–retest volume correlation was r = 0.88 and ICC(1) = 0.86 (95% CI 0.72–0.94); the method took <2 min per image on average.

Methods

  • Dataset: 47 3DUS volumes from singleton pregnancies at 11–14 weeks (28 anterior, 19 posterior placentas); 26 subjects were imaged twice in the same session.
  • Each target image was initialized by manual segmentation of one central 2D slice, propagated to a 3D slab, then segmented via multi-atlas registration and joint label fusion, repeated with bootstrapping and postprocessing.
  • Accuracy was evaluated with leave-one-out atlases against expert manual segmentations using Dice overlap, surface distance, and volume comparison to VOCAL; test–retest reliability used correlation and ICC.

Limitations

  • The paper states that the cohort was limited to singleton first-trimester pregnancies; later stages and nonsingletons remain future work.
  • The paper notes that minimal user input may still be prohibitive for large population studies, and that full automation remains a goal.
  • The paper reports lower posterior performance and attributes it partly to fewer posterior atlases; it also shows sensitivity to manual input extent, with uneven undersegmentation or a thin centerline leading to missing parts or inadequate initialization.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1117/1.jmi.7.1.014004

Cites

21 of 21 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 34 cited sources checked · 0 supported · 0 flagged

  • Findingown data
    § Abstract

    The correlation coefficient between test-retest placental volumes was r = 0.88.

    “The correlation coefficient between test -retest volumes was r ¼ 0.88 , and the intraclass correlation was ICC ð 1 Þ ¼ 0.86 .”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The two 3DUS acquisitions from the same subject within the same session represent the same underlying placental volume, so their volume difference reflects measurement reliability rather than biological change.
  • Findingown data
    § 3.1 Accuracy Experiment

    The mean Dice overlap in the posterior placentas in the accuracy experiment was 0.803 ± 0.055 (N = 19, median 0.821).

    “and 0.803 0.055 in the posterior placentas ( N ¼ 19 , median: 0.821).”
    Quote in the paperSubstantial claim
  • Findingown data
    § 3.1 Accuracy Experiment

    The median Dice overlap in the accuracy experiment was 0.831.

    “The median Dice overlap was 0.831.”
    Quote in the paperSubstantial claim
  • Findingown data
    § Abstract

    The intraclass correlation between test-retest placental volumes was ICC(1) = 0.86.

    “The correlation coefficient between test -retest volumes was r ¼ 0.88 , and the intraclass correlation was ICC ð 1 Þ ¼ 0.86 .”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The two 3DUS acquisitions from the same subject within the same session represent the same underlying placental volume, so their volume difference reflects measurement reliability rather than biological change.
  • Findingown data
    § 3.1 Accuracy Experiment

    The mean Dice overlap in the anterior placentas in the accuracy experiment was 0.838 ± 0.063 (N = 28, median 0.852).

    “The mean Dice overlap was 0.838 0.063 in the anterior placentas ( N ¼ 28 , median: 0.852)”
    Quote in the paperSubstantial claim
  • May combine two claims. 18
  • Classified as probably not a substantial claim. 13

  • May combine two claims.

    Findingown data
    § 2.3.3 Interactive initialization based on manual input on a single 2D slice

    The 3D slab from propagating a single manually segmented 2D slice provides a robust estimation of the 3D shape and position of the placenta but its exact boundaries and extent are not yet identified at this stage.

    “The overlay with the manual 3D segmentation (shown in green) shows that this procedure provides a robust estimation of the 3D shape and position of the placenta but its exact boundaries and extent are not yet identified at this stage.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Expert manual segmentations are accurate enough to serve as the reference standard for evaluating automated segmentation accuracy.

  • May combine two claims.

    Findingown data
    § 3.1 Accuracy Experiment

    For VOCAL, the mean volume was 69.7 ± 29.7 cc, which was significantly smaller than the manual volume (p < 1 × 10^-7, paired t-test).

    “However, for VOCAL, the mean volume was 69.7 29.7 cc , which was significantly smaller than the manual volume ( p &lt; 1 × 10 -7 , paired t -test).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation
    § 1 Introduction

    Using a library of atlases manually created by experts allows the automated segmentation approach to incorporate knowledge on the appearance and shape of the placenta in a diverse set of subjects and leads to highly robust segmentation results in new subjects.

    “Using a library of atlases manually created by experts allows the automated segmentation approach to incorporate knowledge on the appearance and shape of the placenta in a diverse set of subjects and leads to highly robust segmentation results in new subjects.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [15] source

      Not checked in the free run.

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

    Premisecitation
    § 1 Introduction

    Multi-atlas segmentation has been applied successfully to many challenging segmentation problems such as identifying hippocampal subfields in MRI, the spinal cord, and the mitral valve in preoperative 3D echocardiography.

    “This approach has been applied successfully to many challenging segmentation problems 15 such as identifying hippocampal subfields in magnetic resonance images (MRI) of the brain, 17,18 the spinal cord, 19 and the mitral valve in preoperative 3D echocardiography. 20”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [15] source

      Not checked in the free run.

    • cites [17] source

      Not checked in the free run.

    • cites [18] source

      Not checked in the free run.

    • cites [20] source

      Not checked in the free run.

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

    Premisenone given
    § 1 Introduction

    Uterine contractions can dramatically affect the shape of the placenta.

    “Uterine contractions can also dramatically affect the shape of the placenta.”
    Quote in the paperNot a substantial claim

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

    Study design and conductown data
    § 2.2 Expert Manual Segmentation

    Each of the 47 subjects was manually segmented under the supervision of a clinical expert with over 10 years of experience in prenatal ultrasound imaging, who provided guidance, final inspection, and approval.

    “Each of the 47 subjects were manually segmented under the supervision of our clinical expert, who has over 10 years of experience in prenatal ultrasound imaging and who has segmented hundreds of placentas for other research endeavors. The clinical expert provided guidance for the manual segmentation training and performed final inspection and approval for each segmentation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation
    § 1 Introduction

    Current techniques for quantifying placental volume and morphology from 3DUS images require substantial manual input, which can be cumbersome and poorly standardized.

    “As a result of these challenges, current techniques for quantifying placental volume and morphology from 3DUS images require substantial amount of manual input from the user, which can be cumbersome and poorly standardized.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [8] source

      Not checked in the free run.

    • cites [9] source

      Not checked in the free run.

  • May combine two claims.

    Premisereasoning
    § 1 Introduction

    The shape and relative location of the placenta are highly variable, which can make automatic detection of the placenta location difficult.

    “Additionally, the shape and relative location of the placenta are highly variable. In particular, placentas can be positioned anteriorly, posteriorly, or laterally to the amniotic sac. This variability can make it difficult to even detect the location of the placenta automatically, much less determine its precise boundaries.”
    Quote in the paperSubstantial claim

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

    Study design and conductnone given
    § 2.1 Data

    The study protocol was approved by the University of Pennsylvania Institutional Review Board and informed consent was obtained from all subjects.

    “The study protocol was approved by the University of Pennsylvania Institutional Review Board (IRB# 811129). Informed consent was obtained from all subjects.”
    Quote in the paperNot a substantial claim

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

    Premisecitation
    § 1 Introduction

    A first fully automated algorithm based on multi-atlas segmentation techniques was proposed by Oguz et al.

    “A first fully automated algorithm based on multi-atlas segmentation techniques was proposed by Oguz et al., 10”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [10] Oguz, 2016, Fully automated placenta segmentation from 3D ultrasound images

      Not checked in the free run.

  • May combine two claims.

    Premisecitation
    § 1 Introduction

    The previously demonstrated fully automated multi-atlas framework was limited to anterior-only placentas, and its generalization to arbitrarily positioned placentas is problematic due to initialization problems from variability in location and pose.

    “Although we previously demonstrated the applicability of this fully automated framework in a limited dataset of anterior-only placentas, 10 the generalization of this technique to arbitrarily positioned placentas is problematic due to initialization problems arising from the wide variability in location and pose of the placenta.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [10] Oguz, 2016, Fully automated placenta segmentation from 3D ultrasound images

      Not checked in the free run.

  • May combine two claims.

    Premisereasoning
    § 2.3.4 Multi-atlas joint label fusion

    Accurate image registration is of paramount importance to MALF success, and this can be problematic for placental imaging because the shape and position of the placenta within the imaging window are highly variable.

    “Since the MALF is a registration-based method, accurate image registration is of paramount importance to its success. This can be problematic for our application because the shape and position of the placenta within the imaging window are highly variable.”
    Quote in the paperSubstantial claim

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

    Premisecitation
    § 1 Introduction

    Two-dimensional and three-dimensional ultrasound have been used to study in utero placental size and morphology early in pregnancy in relation to birthweight and preeclampsia.

    “In particular, two-dimensional and three-dimensional ultrasound (3DUS) have been used to study in utero placental size and morphology early in pregnancy in relation to birthweight 4 and preeclampsia. 5,6”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [4] source

      Not checked in the free run.

    • cites [5] source

      Not checked in the free run.

    • cites [6] source

      Not checked in the free run.

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

    Premisecitation
    § 1 Introduction

    Alternative methods based on deep learning have been proposed for placental segmentation.

    “and a few alternative methods based on the recently popular deep learning methods have been proposed. 11 -14”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [11] source

      Not checked in the free run.

    • cites [12] source

      Not checked in the free run.

    • cites [13] source

      Not checked in the free run.

    • cites [14] source

      Not checked in the free run.

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

    Premisenone given
    § 1 Introduction

    Posterior placentas are often particularly challenging to segment due to shadowing from the fetus.

    “Furthermore, posterior placentas are often particularly challenging to segment due to shadowing from the fetus.”
    Quote in the paperNot a substantial claim

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

    Premisenone given
    § 2.3.4 Multi-atlas joint label fusion

    The placenta appears bright in 3DUS images, whereas amniotic fluid appears dark.

    “The placenta appears bright in the 3DUS images, whereas the fluid appears dark.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisereasoning
    § 1 Introduction

    Recently developed fully automated methods are becoming increasingly capable for population-based studies but are currently far from guaranteeing adequate performance in every subject; difficult-to-segment images continue to pose a significant challenge.

    “Although these recently developed methods are becoming increasingly capable for fully automated population-based studies, they are currently far from guaranteeing adequate performance in every subject; as such, difficult-to-segment images such as those with a high amount of noise or unusually shaped placenta continue to pose a significant challenge.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given
    § 4.1 Segmentation Accuracy and Comparison to Other Placenta Segmentation Methods

    Such failure cases may form a relatively small portion of a large dataset and can arguably be excluded from a population study, although selection bias may be a concern.

    “Such failure cases may form a relatively small portion of a large dataset and can arguably be excluded from a population study (although selection bias may be a concern).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 4.1 Segmentation Accuracy and Comparison to Other Placenta Segmentation Methods

    The Oguz et al. 14 method combines deep learning techniques and the multi-atlas approach, unlike the currently presented method which only uses the multi-atlas component.

    “This method 14 combines deep learning techniques and the multiatlas approach, unlike the currently presented method which only uses the multi-atlas component.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [14] source

      Not checked in the free run.

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

    Interpretationreasoning
    § 4.2 Performance on Anterior versus Posterior Placentas

    An additional contributor to the performance difference may be the relatively small number of posterior subjects in the atlas library; increasing the number of posterior placentas in the library might provide a better representation of shape and appearance distribution and help further minimize the performance difference.

    “An additional contributor to the performance difference may be the relatively small number of posterior subjects in our atlas library; increasing the number of posterior placentas in the library might provide a better representation of shape and appearance distribution and help further minimize the performance difference.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data
    § 5 Conclusion

    The presented minimally interactive placenta segmentation method achieves excellent performance while only requiring user input on a single 2D slice of a 3DUS volume.

    “We presented a minimally interactive placenta segmentation method that achieves excellent performance while only requiring user input on a single 2D slice of a 3DUS volume.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data
    § 4.3 Sensitivity to Manual Input

    These results show that precise manual input is not strictly required, and that it may be better to err on the side of oversegmentation for initializing this method.

    “These results show that precise manual input is not strictly required, and that it may be better to err on the side of oversegmentation for initializing this method.”
    Quote in the paperSubstantial claim

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

    Premisereasoning
    § 4.1 Segmentation Accuracy and Comparison to Other Placenta Segmentation Methods

    The Looney et al. 13 method is also evaluated against random walker results rather than expert manual segmentations.

    “Like the previous version, 11 this method is also evaluated against random walker results rather than expert manual segmentations.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Looney et al., 13

      Not checked in the free run.

  • May combine two claims.

    Premisecitation
    § 4.1 Segmentation Accuracy and Comparison to Other Placenta Segmentation Methods

    An extension of the multi-atlas approach reported by Oguz et al. 14 makes the method applicable to both anterior and posterior placentas and achieves a mean Dice overlap of 0.86.

    “An extension of the multi-atlas approach, 10 reported by Oguz et al., 14 makes the method applicable to both anterior and posterior placentas and achieves a mean Dice overlap of 0.86.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Oguz et al., 14

      Not checked in the free run.

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

    Premisecitation
    § 4.1 Segmentation Accuracy and Comparison to Other Placenta Segmentation Methods

    Another fully automated deep learning method, proposed by Yang et al. 12, is a general-purpose segmentation technique that is applicable to many different organs.

    “Another fully automated deep learning method, proposed by Yang et al., 12 is a generalpurpose segmentation technique that is applicable to many different organs.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Yang et al., 12

      Not checked in the free run.

25 of 25

  • The paper relies on this without support.

    Expert manual segmentations are accurate enough to serve as the reference standard for evaluating automated segmentation accuracy.

    “Dice overlap and surface distance were used to quantify the automated segmentation accuracy compared to expert manual segmentations.”
    “Each of the 47 subjects were manually segmented under the supervision of our clinical expert, who has over 10 years of experience in prenatal ultrasound imaging and who has segmented hundreds of placentas for other research endeavors.”

  • The paper relies on this without support.

    The two 3DUS acquisitions from the same subject within the same session represent the same underlying placental volume, so their volume difference reflects measurement reliability rather than biological change.

    “26 of these subjects (14 anterior and 12 posterior) were imaged twice within the same session; the patients were allowed to move and the ultrasound probe position was changed between the acquisition of the two images.”

  • The paper relies on this without support.

    Later stages of pregnancy are challenging to image with 3DUS as the placenta becomes too big for the imaging window.

    “Later stages of pregnancy are challenging to image with 3DUS as the placenta becomes too big for the imaging window.”

  • The paper relies on this without support.

    No fully automated method will work 100% of the time.

    “it is well known that no fully automated method will work 100% of the time”

  • The paper relies on this without support.

    Fetal shadowing is common in posterior placentas.

    “where fetal shadowing is common”

  • The paper relies on this without support.

    The dataset used to evaluate the Oguz et al. 10 method is a subset of the one used in the current manuscript.

    “We note that the dataset used to evaluate this method is a subset of the one used in the current manuscript.”

  • The paper relies on this without support.

    The dataset used to evaluate the Oguz et al. 14 method is the same as the one used in the current manuscript.

    “We note that the dataset used to evaluate this method is the same as the one used in the current manuscript.”
7 of 7

Claims other papers make about this one

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

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Reference 20 is described as addressing a different valve

    Minor

    The text cites reference 20 for segmentation of the mitral valve, whereas the cited paper's title identifies the tricuspid aortic valve.

  • Random-walker study is attributed to a different lead author

    Minor

    The discussion attributes reference 9's random-walker method to Stevenson et al., but the paper's reference 9 lists Collins et al. as its authors.

2 of 2

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Oguz, I., Yushkevich, N., Pouch, A. M., Oguz, B. U., Wang, J., Parameshwaran, S., Gee, J. C., Yushkevich, P. A., & Schwartz, N.
Affiliations
  • Vanderbilt Univ.
  • University of Pennsylvania (US)
ORCID
Venue
Journal of Medical Imaging
Publisher
SPIE-Intl Soc Optical Eng
Published
2020-02-22
Volume
7
Issue
01
Pages
1-14
ISSN
  • 2329-4302 (ISSN-L)
  • 2329-4310
DOI
10.1117/1.jmi.7.1.014004
PMC
PMC7035878
OpenAlex
W3007685270
Semantic Scholar
e5e5f2e2296ee786b678b16c766e0c028ca65b54
Cited by
13
References
21
Citations by year
2025: 4 · 2024: 2 · 2023: 3 · 2022: 1 · 2021: 2 · 2020: 1
FWCI
0.68
Type
journal article
Language
en
Open access
Locations
  • Journal of medical imaging (publishedVersion)
  • PubMed Central (publishedVersion)
  • PubMed (publishedVersion)
Funders
  • Nvidia Corporation
  • Google
  • National Institute of Child Health and Human Development (U01 HD087180)
  • National Institute of Biomedical Imaging and Bioengineering (R01 EB017255)
  • National Institute of Neurological Disorders and Stroke (R01 NS094456)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Medical Image Segmentation Techniques
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Physical Sciences › Engineering › Biomedical Engineering › Medical Imaging and Analysis
Keywords
placenta segmentation, placental volume, 3D ultrasound, semi-automatic segmentation, multi-atlas label fusion, early pregnancy, test-retest reliability, fetal growth, VOCAL

About these checks

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

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-10.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-10.

Statistics and arithmetic are recomputed, not judged.

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

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

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