CoChat
Sign Up
Cited by 38Medical Image Segmentation TechniquesArtificial Intelligence in Healthcare and EducationAdvanced Neural Network Applications

Model-Agnostic Method for Thoracic Wall Segmentation in Fetal Ultrasound Videos

Open access
Shozu, Kanto, Komatsu, Masaaki, Sakai, Akira, Komatsu, Reina, Dozen, Ai, Machino, Hidenori, Yasutomi, Suguru, Arakaki, Tatsuya +6 more·2020·Biomolecules
DOI 10.3390/biom10121691PMC PMC7766150
Open this paper in CoChat — free accountOpen PDF

Abstract

The application of segmentation methods to medical imaging has the potential to create novel diagnostic support models. With respect to fetal ultrasound, the thoracic wall is a key structure on the assessment of the chest region for examiners to recognize the relative orientation and size of structures inside the thorax, which are critical components in neonatal prognosis. In this study, to improve the segmentation performance of the thoracic wall in fetal ultrasound videos, we proposed a novel model-agnostic method using deep learning techniques: the Multi-Frame + Cylinder method (MFCY). The Multi-frame method (MF) uses time-series information of ultrasound videos, and the Cylinder method (CY) utilizes the shape of the thoracic wall. To evaluate the achieved improvement, we performed segmentation using five-fold cross-validation on 538 ultrasound frames in the four-chamber view (4CV) of 256 normal cases using U-net and DeepLabv3+. MFCY increased the mean values of the intersection over union (IoU) of thoracic wall segmentation from 0.448 to 0.493 for U-net and from 0.417 to 0.470 for DeepLabv3+. These results demonstrated that MFCY improved the segmentation performance of the thoracic wall in fetal ultrasound videos without altering the network structure. MFCY is expected to facilitate the development of diagnostic support models in fetal ultrasound by providing further accurate segmentation of the thoracic wall.

AI snapshot

Written from the paper’s full text.

TL;DR

The paper proposes MFCY, a model-agnostic method combining Multi-Frame (MF) and Cylinder (CY) ensemble strategies to improve thoracic wall segmentation in fetal ultrasound videos. MFCY increased mean IoU from 0.448 to 0.493 for U-net and from 0.417 to 0.470 for DeepLabv3+ without modifying network structures.

Key findings

  • MFCY improved thoracic wall segmentation mean IoU: 0.448 (U-net) to 0.493 (U-net + MFCY), and 0.417 (DeepLabv3+) to 0.470 (DeepLabv3+ + MFCY).
  • MFCY increased mean Recall: 0.568 (U-net) to 0.738 (U-net + MFCY), and 0.525 (DeepLabv3+) to 0.729 (DeepLabv3+ + MFCY).
  • MFCY decreased mean Precision: 0.679 (U-net) to 0.596 (U-net + MFCY), and 0.675 (DeepLabv3+) to 0.566 (DeepLabv3+ + MFCY).
  • MF was also reported effective for segmentation of the thoracic wall, thoracic cavity, whole thorax, heart, and lung in supplementary experiments.

Methods

  • Design: MFCY integrates MF, which uses neighboring frames from ultrasound videos, and CY, which uses three prediction labels—thoracic wall, thoracic cavity, and whole thorax—based on the cylindrical shape of the thoracic wall.
  • Data: 538 four-chamber-view ultrasound frames from 280 videos of 256 normal cases; singleton pregnant women in the second trimester at four Showa University Hospitals between April 2018 and May 2019; median gestational age 20 weeks (range 18–28 weeks).
  • Analysis: Five-fold cross-validation with U-net and DeepLabv3+; metrics were IoU, Dice, Precision, and Recall; thoracic wall labels were manually annotated, while thoracic cavity and whole thorax labels were automatically generated from the thoracic wall label.

Limitations

  • Stated: MFCY was tested only with U-net and DeepLabv3+, so improvement with other deep learning algorithms is unclear.
  • Stated: Data came from only two ultrasonography machines from a single company, limiting equipment robustness.
  • Stated: The dataset included only normal fetuses, so effectiveness in fetal abnormalities such as congenital heart diseases was not examined; clinical validation was out of scope, and the maximum mIoU of 0.493 may still be relatively low for clinical utility.
  • No additional limitations are evident beyond those stated in the paper.

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

Integrity signals

Registry

Crossref · 10.3390/biom10121691

Cites

39 of 40 references verified1 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 23 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    MFCY increased the mean IoU of thoracic wall segmentation from 0.448 to 0.493 for U-net.

    “MFCY increased the mean values of the intersection over union (IoU) of thoracic wall segmentation from 0.448 to 0.493 for U-net and from 0.417 to 0.470 for DeepLabv3+.”
    Quote in the paperSubstantial claim
  • Findingown data

    MFCY increased the mean IoU of thoracic wall segmentation from 0.417 to 0.470 for DeepLabv3+.

    “MFCY increased the mean values of the intersection over union (IoU) of thoracic wall segmentation from 0.448 to 0.493 for U-net and from 0.417 to 0.470 for DeepLabv3+.”
    Quote in the paperSubstantial claim
  • Findingown data

    The improvement of segmentation results depended on each image.

    “The improvement of segmentation results depended on each image (Figure 3a and Figure 4a).”
    Quote in the paperNot a substantial claim
  • Findingown data

    In the authors' preliminary experiment, the segmentation performance of existing CNN models for the thoracic wall was suboptimal compared to the heart and lung.

    “In fact, our preliminary experiment of the segmentation of the heart, lung, and thoracic wall indicated that the segmentation performance of the existing convolutional neural network (CNN) models corresponding to the thoracic wall were suboptimal compared to the heart and lung (Appendix A Table A1).”
    Quote in the paperSubstantial claim
  • Findingown data

    DeepLabv3+ + MFCY had a lower mPrecision than DeepLabv3+ for thoracic wall segmentation (0.566 vs. 0.675).

    “0.675 (DeepLabv3+) vs. 0.566 (DeepLabv3+ + MFCY)”
    Quote in the paperNot a substantial claim
  • May combine two claims. 12
  • Classified as probably not a substantial claim. 17

  • May combine two claims.

    Findingown data

    MFCY improved segmentation performance without making any modifications to the network or requiring any effort for greater annotation, even though the dataset was limited.

    “Remarkably, we improved segmentation performance without making any modifications to the network or requiring any effort for greater annotation, even though the dataset was limited.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    Both MF and CY often improved segmentation results by compensating for failing areas and connecting the segmentation area closer to the cylindrical shape.

    “Both MF and CY often improved segmentation results by compensating for failing areas and connecting the segmentation area closer to the cylindrical shape.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    The dataset was uniformly divided into five independent subsets that were disjointed and almost equal in cardinality.

    “To this end, the dataset was uniformly divided into five independent subsets such that they were disjointed and almost equal in cardinality.”
    Quote in the paperSubstantial claim

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

    Study design and conductnone given

    The study protocol was approved by the Institutional Review Boards of RIKEN, Fujitsu Ltd., Showa University, and the National Cancer Center.

    “The study protocol was approved by the Institutional Review Boards of RIKEN, Fujitsu Ltd., Showa University, and the National Cancer Center (approval ID: Wako1 29–4).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    By referring to the thoracic wall, examiners recognize the relative orientation and size of structures inside the thorax, which allows the detection of congenital abnormalities.

    “By referring to the thoracic wall, examiners ensure that the obtained 4CV is in the appropriate cross-section [11], and then recognize the relative orientation and size of structures inside the thorax, which allows the detection of congenital abnormalities [12,13].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [12] (2013). The International Society of Ultrasound in Obstetrics ISUOG Practice Guidelines (updated): Sonographic screening examination of the fetal heart. Ultrasound Obstet. Gynecol., 41, 348–359. source

      Not checked in the free run.

    • cites [13] Satomi, 2015, Guidelines for fetal echocardiography, Pediatr. Int. source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Ensemble learning improves aggregate performance by combining predictions obtained from multiple trained models.

    “Ensemble learning is a powerful method that improves aggregate performance by combining the predictions obtained from multiple trained models [28].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [28] Sagi, 2018, Ensemble learning: A survey, WIREs Data Min. Knowl. Discov.

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    The thoracic wall labels were manually annotated in the images by an obstetrician.

    “The thoracic wall labels were manually annotated in these images by an obstetrician.”
    Quote in the paperSubstantial claim

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

    Premisecitation

    DeepLabv3+ demonstrated state-of-the-art performance on cityscape images.

    “Chen et al. developed DeepLabv3+ by combining pyramidal pooling modules with an encoder-decoder model and demonstrated its state-of-the-art performance on cityscape images [16].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] Chen, 2018, Encoder-decoder with atrous separable convolution for semantic image segmentation, Lect. Notes Comput. Sci. source

      Not checked in the free run.

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

    Premisecitation

    U-net achieved more accurate segmentation using smaller amounts of training data compared with other methods.

    “Ronneberger et al. developed U-net, based on a fully convolutional network, and achieved more accurate segmentation using smaller amounts of training data compared with the other methods [14].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [14] Ronneberger, 2015, U-net: Convolutional networks for biomedical image segmentation, Lect. Notes Comput. Sci. source

      Not checked in the free run.

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

    Premisecitation

    U-net is particularly suitable for biomedical image segmentation.

    “U-net is particularly suitable for biomedical image segmentation.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [14] Ronneberger, 2015, U-net: Convolutional networks for biomedical image segmentation, Lect. Notes Comput. Sci. source

      Not checked in the free run.

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

    Premisecitation

    The thoracic wall surrounds the thoracic cavity and consists of multiple components, including the thoracic vertebra, ribs, sternum, and muscles.

    “The thoracic wall, which surrounds the thoracic cavity and consists of multiple components, including the thoracic vertebra, ribs, sternum, and muscles [10], is a key structure to evaluate the 4CV frame (Supplementary Figure S1).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [10] Roberts, K.P., and Weinhaus, A.J. (2005). Anatomy of the Thoracic Wall, Pulmonary Cavities, and Mediastinum. Handbook of Cardiac Anatomy, Physiology, and Devices, Humana Press. source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    Data augmentation or fine-tuning were not conducted on U-net and DeepLabv3+ to ensure equal operating conditions for all candidate methods.

    “During the segmentation process, data augmentation or fine-tuning were not conducted on U-net and DeepLabv3+ to ensure equal operating conditions for all candidate methods.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Additional annotation cost for CY is free because the thoracic cavity and whole thorax labels can be automatically generated from the thoracic wall labels.

    “We need two additional labels (thoracic cavity and whole thorax) for CY, but we can automatically generate these two labels from the thoracic wall labels. This means additional annotation cost for CY is free.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    MF is effective for providing further accurate segmentation of not only the thoracic wall, thoracic cavity, and whole thorax, but also the heart and lung.

    “Our results showed that MF is effective to provide further accurate segmentation for not only the thoracic wall, thoracic cavity, and whole thorax, but also the heart and lung.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductown data

    Five-fold cross-validation was used to generalize the performances of the existing models and MF, CY, and MFCY.

    “We utilized five-fold cross-validation to generalize the performances of the existing models and our three methods: MF, CY, and MFCY.”
    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

    All methods were carried out in accordance with the Declaration of Helsinki and the Ethical Guidelines for Medical and Health Research Involving Human Subjects.

    “All methods were carried out in accordance with the declaration of Helsinki and the Ethical Guidelines for Medical and Health Research Involving Human Subjects.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    The appearance of the thoracic wall varies in fetal ultrasounds because it is a mixture of structures of diverse brightness, and artifacts frequently appear in the thoracic wall, caused by the wall itself or the surrounding structures.

    “The appearance of the thoracic wall varies in fetal ultrasounds, because it is a mixture of structures of diverse brightness, and artifacts frequently appear in the thoracic wall, which are caused by the wall itself or the surrounding structures.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    Validation in clinical settings was out of scope in this study; the maximum mIoU (U-net + MFCY) was 0.493, which might be still relatively low for clinical utility.

    “Finally, the validation in clinical settings was out of scope in this study; the maximum mIoU (U-net + MFCY) was 0.493, which might be still relatively low for a clinical utility.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    A small thoracic circumference is present in cases of skeletal dysplasia with narrow thorax.

    “a small thoracic circumference is present in cases of skeletal dysplasia with narrow thorax [33]”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [33] Ruano, 2004, Prenatal diagnosis of fetal skeletal dysplasias by combining two-dimensional and three-dimensional ultrasound and intrauterine three-dimensional helical computer tomography, Ultrasound Obstet. Gynecol. source

      Not checked in the free run.

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

    Premisecitation

    The thoracic wall itself is deformed in cases of pectus excavatum.

    “the thoracic wall itself is deformed in cases of pectus excavatum [34]”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [34] Sucu, 2016, Prenatal diagnosis of pectus excavatum, Turkish J. Obstet. Gynecol. source

      Not checked in the free run.

  • May combine two claims.

    Study design and conductown data

    The dataset in this study included ultrasound videos of only normal fetuses, so the effectiveness of MFCY in cases of fetal abnormalities such as CHDs was not examined.

    “Third, the dataset in this study included ultrasound videos of only normal fetuses, which means that we have not examined the effectiveness of MFCY in the cases of fetal abnormalities such as CHDs.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Since MFCY is a model-agnostic method, it is formally applicable to all deep learning algorithms.

    “Since MFCY is a model-agnostic method, it is formally applicable to all deep learning algorithms.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    Based on the prior knowledge that the thoracic wall has a cylindrical structure, multiple labels (thoracic cavity label and whole thorax label) can be generated from a single label (thoracic wall label).

    “Based on the prior knowledge that the thoracic wall has a cylindrical structure, multiple labels (thoracic cavity label and whole thorax label) can be generated from a single label (thoracic wall label).”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The thoracic wall has a cylindrical structure.

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

    Premisecitation

    Examiners validate that their scanning view is in the appropriate cross-section by obtaining the symmetrical thoracic wall, which has a full length of a rib in each side.

    “examiners validate that their scanning view is in the appropriate cross-section by obtaining the symmetrical thoracic wall, which has a full length of a rib in each side [11].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [11] Bethune, 2013, A pictorial guide for the second trimester ultrasound, Australas. J. Ultrasound Med. source

      Not checked in the free run.

  • May combine two claims.

    Study design and conductown data

    The input data were acquired from only two types of ultrasonography machines operated by a single company, which does not ensure the robustness of the equipment.

    “Second, the input data were acquired from only two types of ultrasonography machines operated by a single company, which does not ensure the robustness of the equipment.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Although the size of the dataset was limited in this study, the methods are considered effective from these qualitative analyses.

    “Although the size of the dataset was limited in this study, it is considered that our methods are effective from these qualitative analyses”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisenone given

    MFCY is based on ensemble learning of the independent predictions of the two methods; it complements time-series information of ultrasound videos and the shape information of the thoracic wall with each other.

    “MFCY is based on ensemble learning of the independent predictions of the two methods; it complements time-series information of ultrasound videos and the shape information of the thoracic wall each other.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    Accurate segmentation of the fetal thoracic wall is essential to construct AI-based models for supporting the assessment of 4CV.

    “Accurate segmentation of the fetal thoracic wall is essential to construct AI-based models for supporting the assessment of 4CV.”
    Quote in the paperNot a substantial claim
28 of 28

  • The authors state this as an assumption.

    The thoracic wall is always cylindrical.

    “CY utilizes the prior knowledge that the thoracic wall is always cylindrical.”
    “the other is the property that the thoracic wall is always cylindrical.”

  • The authors state this as an assumption.

    Neighboring images in an ultrasound video are very similar to each other.

    “one is the property that the ultrasound video is a series of images where neighboring images are very similar to each other”

  • The authors state this as an assumption.

    No prior study has focused on thoracic wall segmentation in fetal ultrasound.

    “In this study, to the best of our knowledge, we attempted for the first time to focus on the thoracic wall segmentation in fetal ultrasound.”

  • The paper relies on this without support.

    The thoracic wall has a cylindrical structure.

    “Based on the prior knowledge that the thoracic wall has a cylindrical structure”

  • The paper relies on this without support.

    The thoracic wall is not a simply connected topological space.

    “the thoracic wall is not a simply connected topological space”
5 of 5

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?

  • Two studies focused on automated detection of fetal facial standard planes.

    “Two studies focused on automated detection of fetal facial standard planes.118,119”

    From Artificial intelligence in obstetric ultrasound: A scoping review. (2023) · cites this paper as [118]

    The Methods (section 2.1) describe segmentation of the fetal thoracic wall in four-chamber-view ultrasound images, not detection of fetal facial standard planes. The Conclusions (section 5) likewise identify improved thoracic-wall segmentation as the study’s result. Page numbers are unavailable in our copy. (LLM, conf=0.90)

    Checked by a reasoning model — may be wrong.

    Experimental — accuracy not yet measured

  • Shozu et al. presented an AI model for the identification of the thoracic wall, which enabled plane detection for 4CV, but also allowed for the detection of thoracic malformations.

    “and Shozu et al. presented an AI model for the identification of the thoracic wall, which enabled plane detection for 4CV, but also allowed for the detection of thoracic malformations [143].”

    From Evolving the Era of 5D Ultrasound? A Systematic Literature Review on the Applications for Artificial Intelligence Ultrasound Imaging in Obstetrics and Gynecology (2023) · cites this paper as [143]

    The Introduction and Discussion (page not provided) describe Shozu et al.’s thoracic-wall segmentation method and its potential to support assessment of the four-chamber view (4CV) and diagnosis of congenital abnormalities. They do not report that the model performed 4CV plane detection or detected thoracic malformations; the claim presents proposed uses as achieved capabilities. (LLM, conf=0.88)

    Experimental — accuracy not yet measured

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.

  • Frame interval attributed to CY rather than MF

    Minor

    The configuration assigns a neighboring-frame interval to CY, but CY is defined using a single image. The neighboring-frame interval is a parameter of MF.

1 of 1

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. Oprea, 2018, A survey on deep learning techniques for image and video semantic segmentation, Appl. Soft Comput. J., 70, 41, 10.1016/j.asoc.2018.05.018 · DOI

Cited by

Sorted by citations

Related

Authors
Shozu, K., Komatsu, M., Sakai, A., Komatsu, R., Dozen, A., Machino, H., Yasutomi, S., Arakaki, T., Asada, K., Kaneko, S., Matsuoka, R., Nakashima, A., Sekizawa, A., & Hamamoto, R.
Affiliations
  • University of Toyama (JP)
  • RIKEN Center for Advanced Intelligence Project (JP)
  • Tokyo Medical and Dental University (JP)
  • SHOWA Medical University (JP)
  • National Cancer Center Research Institute (JP)
  • Fujitsu (Japan) (JP)
Corresponding author
  • Komatsu, Masaaki
  • Hamamoto, Ryuji
ORCID
Venue
Biomolecules
Publisher
MDPI AG
Published
2020-12-17
Volume
10
Issue
12
Pages
1691
ISSN
2218-273X (ISSN-L)
DOI
10.3390/biom10121691
PMC
PMC7766150
OpenAlex
W3110883886
Cited by
38
References
39
Citations by year
2025: 5 · 2024: 7 · 2023: 7 · 2022: 8 · 2021: 11
FWCI
2.53
Type
journal article
Language
en
Open access
Licence
Locations
  • Biomolecules (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Ministry of Education, Culture, Sports, Science and Technology (the subsidy for the Advanced Integrated Intelligence Platform)
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 › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
Keywords
prenatal ultrasound, four-chamber view, neonatal prognosis, U-Net, DeepLabV3+
MeSH terms
Machine Learning, Algorithms, Artificial Intelligence, Heart, Humans, Image Processing, Computer-Assisted, Prenatal Diagnosis, Prognosis, Models, Statistical, Ultrasonography, Prenatal, Neural Networks, Computer, Computational Biology, Thoracic Wall

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