CoChat
Sign Up
Cited by 18Medical Image Segmentation TechniquesArtificial Intelligence in Healthcare and EducationCongenital Heart Disease Studies

Convolutional Neural Networks for Automated Fetal Cardiac Assessment using 4D B-Mode Ultrasound

Open access
Philip, Manna E., Sowmya, Arcot, Avnet, Hagai, Ferreira, Ana, Stevenson, Gordon, Welsh, Alec·2019·2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)
DOI 10.1109/isbi.2019.8759377
Open this paper in CoChat — free accountOpen PDF

Abstract

Structural and functional assessment of the fetal heart is currently performed using ultrasound-based imaging biomarkers that require annotation by a trained observer. This requires experience and expertise and is subject to observer variability. In this work, a Convolutional Neural Network (CNN) was implemented for segmentation of the fetal annulus in 250 cases to provide automated measurements of the excursion of the mitral and tricuspid valve annular planes (TAPSE/MAPSE). These were evaluated against manual annotation. After training, the CNN achieved a Dice score of 0.78 (±0.02;σ) and the excursion measures had an RMSE of <; 0.16 cm. Results show the feasibility of using a CNN to detect the fetal annuli and measure TAPSE/MAPSE for cardiac functional assessment.

AI snapshot

Written from the paper’s full text.

TL;DR

This paper implements a U-net CNN to segment fetal mitral and tricuspid annuli from 4D STIC B-mode ultrasound and automatically compute TAPSE/MAPSE. It reports feasibility, with good segmentation for high-quality data but weaker automated MAPSE correlation than expert agreement.

Key findings

  • The CNN achieved Dice scores of 0.78 for tricuspid annulus (TA) segmentation and 0.77 for mitral annulus (MA) segmentation; for volumes with quality score ≥ 8, Dice was 0.85 for both TA and MA.
  • Automated TAPSE measurement had correlation coefficient r = 0.61 with reference and RMSE = 0.14 cm; expert-to-expert TAPSE correlation was r = 0.89.
  • Automated MAPSE measurement had correlation coefficient r = 0.30 with reference and RMSE = 0.18 cm; expert-to-expert MAPSE correlation was r = 0.77.
  • TAPSE/MAPSE were calculated for 24 randomly chosen volumes; outliers were removed using median absolute deviation × 2.5, discarding measurements greater than 1.13 cm for TAPSE and 0.74 cm for MAPSE, removing 2 and 5 volumes respectively.

Methods

  • Prospective observational study approved by ethics committees; participants > 18 years old with singleton, healthy pregnancy were recruited. A total of 295 fetal heart 4D STIC volumes were acquired at 4-chamber view from 95 participants on a GE Voluson E8 with a RAB 6-D 3D transducer; mean gestational age was 30.7 weeks (range 22.9–38.0).
  • Volumes were quality-scored out of 10; only scores ≥ 6 were used, yielding 150 4D volumes. Manual labelling was performed on 250 3D volumes: the first frame of the 150 chosen volumes plus the 20th frame of a randomly chosen 100 of those 150. Data were split 80:10:10 for training, test, and validation, with augmentation including flips, rotation at ±5° and ±10° in the y-direction, and added speckle noise.
  • A U-net architecture was implemented in Keras with TensorFlow, trained on two NVIDIA GeForce GTX 1080 GPUs, using Adam, Dice loss, 100 epochs, batch size 5, and 10% validation split. Two separate networks were used for MA and TA segmentation. TAPSE/MAPSE were computed from the z-plane centroid displacement of the annulus segmentation, without manually identifying end-systolic and end-diastolic frames; Bland-Altman plots, correlation coefficients, and RMSE were used for evaluation.

Limitations

  • The paper states that fetal movements can shift valve orientation over the cardiac cycle, and the naïve method does not account for this when computing planar excursion. It also states that some outliers came from five-chamber-view acquisitions including the aorta, causing large segmentation error, and that fetal position/apex rotation is another source of error.
  • The paper states that limited pre- and post-processing was performed, that the clinical acquisition system has since been superseded, and that temporal information in the 4D data was not used; it suggests LSTM exploration and a multi-class CNN instead of two separate binary segmentation networks.
  • The text makes evident that frames from a single patient may appear in both training and test sets, though the authors state bias chances were reduced because first and 20th frames were ensured to be at different cardiac phases. The dataset was also considered small enough that data augmentation was needed to train the network.

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

Integrity signals

Registry

Crossref · 10.1109/isbi.2019.8759377

Cites

12 of 12 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 8 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    The root mean squared error between automated and reference MAPSE measurements was 0.18 cm.

    “The r for automated MAPSE measurement and reference was 0.30 and between experts was 0.77 and the RMSE was 0.18.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The two frames with the largest displacement from the centroid of the first frame in the positive/negative direction correspond to the end-diastolic and end-systolic frames.
    • The authors state this as an assumption: The displacement in the z-plane of the centroid (cm) is the annular excursion value (TAPSE/MAPSE).
  • Findingreasoning

    For some subjects, the automated method over-estimates displacement compared to human measurement.

    “Results show for some subjects the method over-estimates displacement compared to human measurement.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The two frames with the largest displacement from the centroid of the first frame in the positive/negative direction correspond to the end-diastolic and end-systolic frames.
    • The authors state this as an assumption: The displacement in the z-plane of the centroid (cm) is the annular excursion value (TAPSE/MAPSE).
  • Findingown data

    The CNN achieved a Dice similarity coefficient of 0.78 for tricuspid annulus (TA) segmentation.

    “Dice Similarity Co-efficient values of 0.78 for TA segmentation and 0.77 for MA segmentation were achieved.”
    Quote in the paperSubstantial claim
  • Findingown data

    The automated TAPSE measurement had a correlation coefficient r = 0.61 with the reference measurement.

    “For the automated TAPSE measurement we obtained a correlation coefficient r = 0.61, while that between experts was r = 0.89.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The two frames with the largest displacement from the centroid of the first frame in the positive/negative direction correspond to the end-diastolic and end-systolic frames.
    • The authors state this as an assumption: The displacement in the z-plane of the centroid (cm) is the annular excursion value (TAPSE/MAPSE).
  • Findingnone given

    TAPSE/MAPSE values for good quality data were better than those with low quality scores.

    “The TAPSE/MAPSE values for good quality data were better than those that had low quality scores.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: The two frames with the largest displacement from the centroid of the first frame in the positive/negative direction correspond to the end-diastolic and end-systolic frames.
    • The authors state this as an assumption: The displacement in the z-plane of the centroid (cm) is the annular excursion value (TAPSE/MAPSE).
  • Classified as probably not a substantial claim. 13
  • May combine two claims. 8

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

    Study design and conductnone given

    Batch normalization and dropout were added to avoid over-fitting.

    “The only changes were addition of the 3D layers to 2D and the addition of batch normalization and dropout to avoid over-fitting.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Manual labelling was performed on 250 3D volumes based on an annulus segmentation protocol designed by experts to ensure labelling uniformity.

    “Manual labelling was performed on 250 3D volumes based on an annulus segmentation protocol designed by experts to ensure labelling uniformity.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    The automated tool can potentially reduce operator variability and reduce the time required to measure TAPSE/MAPSE.

    “This work has shown that automation of TAPSE and MAPSE measurements is possible which is important as it points to augmentation of this method to develop a tool that can reduce operator variability and reduce the time required to measure TAPSE/MAPSE.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The two frames with the largest displacement from the centroid of the first frame in the positive/negative direction correspond to the end-diastolic and end-systolic frames.
    • The authors state this as an assumption: The displacement in the z-plane of the centroid (cm) is the annular excursion value (TAPSE/MAPSE).

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

    Premisecitation

    CNNs have found success in many computer vision tasks and are now being applied by research groups for segmentation of the heart.

    “CNN have found success in many computer vision tasks and are now being applied by research groups for segmentation of the heart.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [4] source

      Not checked in the free run.

    • cites [5]

      Not checked in the free run.

    • cites [6] source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    A quality assurance scoring system was designed in consultation with experts to reduce the performance influence of differing operators.

    “To reduce performance influence of differing operators, a quality assurance scoring system was designed in consultation with the experts.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    The correlation coefficient for MAPSE is lower than that for TAPSE for both experts and the proposed automated method, mainly owing to rotational movement of the mitral annulus due to circular fiber orientation in the left ventricle.

    “It is observed that the correlation coefficient for experts as well as proposed method for MAPSE is lower than those for TAPSE, mainly owing to the rotational movement of the MA due to the circular fiber orientation in left ventricle, in addition to the planar excursion which makes the MAPSE measurement more challenging than TAPSE.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The two frames with the largest displacement from the centroid of the first frame in the positive/negative direction correspond to the end-diastolic and end-systolic frames.
    • The authors state this as an assumption: The displacement in the z-plane of the centroid (cm) is the annular excursion value (TAPSE/MAPSE).

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

    Study design and conductcitation

    Outliers were removed using a median absolute deviation (MAD) * 2.5 cut-off, discarding measurements greater than 1.13 cm for TAPSE and 0.74 cm for MAPSE.

    “Of these, outliers were removed using the median absolute deviation (MAD) * 2.5 to provide a cut-off as previously described [12].Measurements greater than 1.13cm and 0.74cm for TAPSE and MAPSE were discarded, which removed 2 and 5 data volumes, from each group respectively.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [12] source

      Not checked in the free run.

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

    Study design and conductnone given

    Limited pre- and post-processing was performed on the data.

    “Limited pre-and post-processing was performed to the data which shows the potential for this technique with technical improvements and improved testing.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductnone given

    Only volumes with a quality score of 6 or above were considered for this study, and 150 4D volumes satisfied this criterion.

    “Only volumes with a score ≥ of 6 and above were considered for this study.150 4D volumes satisfied the above criteria.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    The study did not take into account the temporal information available in the 4D data.

    “In this study, we do not take into account the temporal information available in the 4D data, which can be explored using LSTM's to possibly improve performance.”
    Quote in the paperSubstantial claim

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

    Study design and conductreasoning

    Frames from a single patient may be in both training and test sets, but chances of bias were reduced because the first and the 20th frame were ensured to be at different phases of the cardiac cycle.

    “Frames from a single patient may be in both training and test sets, but chances of bias were reduced as the first and the 20 th frame were ensured to be at different phases of the cardiac cycle.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    Data augmentation methods were used because the data set was considered small to train a network from scratch.

    “As the size of the data set was considered small to train a network from scratch, data augmentation methods were used to create more samples, using flip up/down, flip left/right and rotation at ±5 • and ±10 • in the y-direction.”
    Quote in the paperSubstantial claim

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

    Premisecitation

    TAPSE reflects longitudinal function and is an indicator of disease progression.

    “This distance has been shown to reflect longitudinal function and be an indicator of disease progression [2].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [2] source

      Not checked in the free run.

  • May combine two claims.

    Premisereasoning

    Manual TAPSE/MAPSE measurement stages introduce potential error and require time, pointing to the need for automation of the measurement of these imaging biomarkers.

    “All these stages introduce potential error and require time, which points to the need for automation of the measurement of these imaging biomarkers.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Manual TAPSE/MAPSE measurement is subject to observer variability.

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

    Premisecitation

    A semi-automated algorithm based on cross correlation to track the tricuspid annulus points has been proposed to obtain TAPSE.

    “A semi-automated algorithm has been proposed [9], based on cross correlation to track the tricuspid annulus (TA) points to obtain the TAPSE.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [9] source

      Not checked in the free run.

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

    Premisecitation

    Automated mitral and tricuspid annular excursion have been explored in adult hearts using additional M-mode and ECG information.

    “Automated mitral and tricuspid annular excursion have been explored in adult hearts by Storve et al. [7,8] using additional M-mode and ECG information to obtain the results.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [7] source

      Not checked in the free run.

    • cites [8] source

      Not checked in the free run.

16 of 16

  • The paper relies on this without support.

    Manual TAPSE/MAPSE measurement is subject to observer variability.

    “This requires experience and expertise and is subject to observer variability.”

  • The authors state this as an assumption.

    The two frames with the largest displacement from the centroid of the first frame in the positive/negative direction correspond to the end-diastolic and end-systolic frames.

    “The two frames with the largest displacement from the centroid of the first frame in the positive / negative direction were considered to correspond to the ED / ES frame.”

  • The authors state this as an assumption.

    The displacement in the z-plane of the centroid (cm) is the annular excursion value (TAPSE/MAPSE).

    “The displacement in the z-plane of the centroid (cm) was considered the particular annular excursion value (TAPSE/MAPSE).”
3 of 3

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.

  • Reported training and test counts do not match the stated split

    Major

    An 80:10:10 split of the 250 labelled volumes would give 200 training, 25 test and 25 validation volumes. The results instead say 150 manually annotated samples were used for training and 100 volumes for testing.

  • Abstract RMSE bound excludes the reported MAPSE result

    Minor

    The abstract says the excursion measurements had an RMSE below 0.16 cm, but the reported MAPSE RMSE is 0.18 cm.

2 of 2

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Philip, M. E., Sowmya, A., Avnet, H., Ferreira, A., Stevenson, G., & Welsh, A.
Affiliations
  • UNSW Sydney (AU)
  • Sheba Medical Center (IL)
  • Royal Hospital for Women (AU)
ORCID
Venue
2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)
Publisher
IEEE
Published
2019-04
Pages
824-828
DOI
10.1109/isbi.2019.8759377
OpenAlex
W2958958034
Cited by
18
References
12
Citations by year
2025: 2 · 2024: 6 · 2023: 4 · 2022: 3 · 2021: 1 · 2020: 2
FWCI
0.48
Type
conference paper
Language
en
Open access
Licence
Locations
  • IEEE International Symposium on Biomedical Imaging (ISBI) (publishedVersion)
  • UNSWorks (University of New South Wales, Sydney, Australia) (submittedVersion, cc-by-nc-nd)
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
  • Health Sciences › Medicine › Epidemiology › Congenital Heart Disease Studies
Keywords
convolutional neural networks, tricuspid annular plane systolic excursion, mitral annulus, tricuspid annulus, observer variability

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