CoChat
Sign Up
Cited by 11Medical Image Segmentation TechniquesAdvanced Neural Network ApplicationsRadiomics and Machine Learning in Medical Imaging

Automated placental abruption identification using semantic segmentation, quantitative features, SVM, ensemble and multi-path CNN

Open access
Asadpour, Vahid, Puttock, Eric J., Getahun, Darios, Fassett, Michael J., Xie, Fagen·2023·Heliyon
DOI 10.1016/j.heliyon.2023.e13577PMC PMC9957707
Open this paper in CoChat — free account

Abstract

The placenta is a fundamental organ throughout the pregnancy and the fetus' health is closely related to its proper function. Because of the importance of the placenta, any suspicious placental conditions require ultrasound image investigation. We propose an automated method for processing fetal ultrasonography images to identify placental abruption using machine learning methods in this paper. The placental imaging characteristics are used as the semantic identifiers of the region of the placenta compared with the amniotic fluid and hard organs. The quantitative feature extraction is applied to the automatically identified placental regions to assign a vector of optical features to each ultrasonographic image. In the first classification step, two methods of kernel-based Support Vector Machine (SVM) and decision tree Ensemble classifier are elaborated and compared for identification of the abruption cases and controls. The Recursive Feature Elimination (RFE) is applied for optimizing the feature vector elements for the best performance of each classifier. In the second step, the deep learning classifiers of multi-path ResNet-50 and Inception-V3 are used in combination with RFE. The resulting performances of the algorithms are compared together to reveal the best classification method for the identification of the abruption status. The best results were achieved for optimized ResNet-50 with an accuracy of 82.88% ± SD 1.42% in the identification of placental abruption on the testing dataset. These results show it is possible to construct an automated analysis method with affordable performance for the detection of placental abruption based on ultrasound images.

AI snapshot

Written from the paper’s full text.

TL;DR

The paper develops an automated ultrasound-image pipeline that segments the placenta, extracts radiomic features, and classifies placental abruption using SVM, AdaBoost ensemble, and multi-path CNN models. The best held-out test accuracy reported was 82.88% ± SD 1.42% for optimized MP-ResNet-50.

Key findings

  • On the test dataset, MP-ResNet-50 achieved the highest accuracy of 82.88% ± SD 1.42%, with sensitivity 82.17% ± SD 2.74%, specificity 83.66% ± SD 1.99%, PPV 84.71% ± SD 1.49%, and AUC 88.31% ± SD 1.18%; MP-Inception-V3 achieved 81.80% ± SD 1.72% accuracy.
  • On the training dataset, the selected-feature image-level AdaBoost ensemble reached the best accuracy of 83.24% and AUC of 92.24%, while SVM reached 81.22% accuracy and AUC 83.04%.
  • Session-level analysis underperformed image-level analysis: selected-feature AdaBoost ensemble reached 80.12% accuracy and AUC 88.87%, and SVM reached 78.06% accuracy and AUC 78.99% on training.
  • On the test dataset, SVM and ensemble classifiers were lower than CNNs: image-level AdaBoost ensemble reached 76.96% accuracy and AUC 81.44%, while image-level SVM reached 70.68% accuracy and AUC 76.73%.

Methods

  • Retrospective ultrasound dataset from Kaiser Permanente Southern California medical centers, January 1, 2018 to December 31, 2018: 645 case images from 97 sessions/61 pregnancies and 453 control images from 97 sessions/66 pregnancies; 80% of sessions were used for training and 20% for testing.
  • Pipeline included image warping to 1280 × 960, gray-scale normalization, Daubechies wavelet filtering, semantic segmentation with a guided filter, extraction of 104 quantitative radiomic features from six groups, PCA for session-level feature merging, RFE feature selection, and Bayesian hyperparameter optimization.
  • Classifiers compared were kernel-based SVM, AdaBoost decision-tree ensemble, and multi-path ResNet-50 and Inception-V3 CNNs fusing image, wavelet-filtered image, and quantitative-feature channels; five-fold cross-validation was used for all classifiers.

Limitations

  • The paper states that test performance showed underperformance relative to training regardless of classifier, potentially due to overfitting, and that future work may use 3D ultrasonography to improve performance.
  • The paper states that the data used are confidential and not available.
  • Evident from the text: the study uses a single-year, single-health-system retrospective sample, manually reviewed/selected images, no external validation, and no abruption staging available at the time of imaging.

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

Integrity signals

Registry

Crossref · 10.1016/j.heliyon.2023.e13577

Cites

23 of 23 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 14 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    The best results were achieved for optimized ResNet-50 with an accuracy of 82.88% ± SD 1.42% in the identification of placental abruption on the testing dataset.

    “The best results were achieved for optimized ResNet-50 with an accuracy of 82.88% ± SD 1.42% in the identification of placental abruption on the testing dataset.”
    Quote in the paperSubstantial claim
  • Findingnone given

    When each optimized model is applied to the test dataset, the performance shows some underperformance regardless of the classifiers.

    “when each optimized model is applied to the test dataset, the performance shows some underperformance regardless of the classifiers”
    Quote in the paperSubstantial claim
  • Findingnone given

    The performance of optimized models at the individual image level is superior to the optimized models at the session level.

    “the performance of optimized models at the individual image level is superior to the optimized models at the session level”
    Quote in the paperSubstantial claim
  • Findingreasoning

    The results for session level analysis show underperformance compared to individual image analysis in SVM and Ensemble classifiers.

    “The results for session level analysis shows under performance compared to individual image analysis in SVM and Ensemble classifiers.”
    Quote in the paperNot a substantial claim
  • Findingown data

    The performance of the two CNN classifiers is better than those of SVM and Ensemble classifiers.

    “The performance of these two CNN classifiers is better than those of SVM and Ensemble classifiers.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 23
  • May combine two claims. 16
  • The quote only partly states this claim. 2

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

    Findingnone given

    The high-ranked quantitative features that survive the RFE elimination process may not directly be related to the classification outcome, but the subset of the features improves the classifier performance.

    “The high-ranked quantitative features that survive the elimination process may not directly be related to the classification outcome, however, the subset of the features improves the classifier performance.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The MP-ResNet-50 Classifier led to the highest accuracy of 82.88% and PPV of 84.71% on the test dataset.

    “The MP-ResNet-50 Classifier led to the highest accuracy of 82.88% and PPV of 84.71%.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    The performance of the CNN classifiers has a small reduction and overcomes SVM and ensemble classifiers significantly on the test dataset.

    “the performance of the CNN classifiers has a small reduction and overcomes SVM and ensemble classifiers significantly”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The Inception-V3 resulted in the highest accuracy while using 27 features only, while the SVM and the decision tree ensemble reached lower accuracies with more than twice the number of selected features.

    “The Highest performance model, the Inception-V3, resulted in the highest accuracy while using 27 features only, however, the SVM and the decision tree ensemble reached lower accuracies with more than twice the number of selected features.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The accuracy reached 95% of maximal performance value at 52 features for SVM and 72 features for Ensemble.

    “The accuracy reached 95% of maximal performance value at 52 features for SVM and 72 for Ensemble, respectively.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingreasoning

    The AUC has a reduction of 3.11% and accuracy has a reduction of 2.59% compared to the individual image analysis.

    “It could be observed that the AUC has a reduction of 3.11% and accuracy has a reduction of 2.59% compared to the individual image analysis.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    The decision tree with AdaBoost Ensemble resulted in an AUC of 88.87% and an accuracy of 80.12% for the full length of the feature vector at session level.

    “The decision tree with AdaBoost Ensemble resulted in an AUC of 88.87% and an accuracy of 80.12% for the full length of the feature vector.”
    Quote in the paperNot a substantial claim

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

    Findingown data

    The tuning process for SVM achieved the best model at iteration 158 and the Ensemble model reached the optimized fitness at iteration 51.

    “The tuning process for SVM achieved the best model at iteration 158 and the Ensemble model reached the optimized fitness at iteration 51.”
    Quote in the paperNot a substantial claim

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

    Findingown data

    The RFE feature optimization led to 54 features for MP-ResNet-50 and 27 features for Inception-V3.

    “The RFE feature optimization is used for these networks that led to 54 features for MP-ResNet-50 and 27 features for Inception-V3.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The observation with 30 repetitions resulted in 85.95% ± SD 1.16% accuracy for MP-ResNet-50 and 87.68% ± SD 0.75% for Inception-V3.

    “The observation with 30 repetitions resulted in 85.95% ± SD 1.16% accuracy for MP-ResNet-50 and 87.68% ± SD 0.75% for Inception-V3.”
    Quote in the paperSubstantial claim

  • May combine two claims. The quote only partly states this claim.

    Findingown data

    The decision tree with AdaBoost Ensemble reached the best AUC of 92.24% and the best accuracy of 83.24% on the training dataset with selected features at image level.

    “It could be observed that the decision tree with AdaBoost Ensemble reached the best AUC of 92.24% and the best accuracy of 83.24%.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Placental abruption is a serious complication of pregnancy that contributes to both maternal and infant morbidity and mortality.

    “It is a serious complication of pregnancy that contributes to both maternal and infant morbidity and mortality [2,3].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [2] Ananth, 2017

      Not checked in the free run.

    • cites [3] Glantz, 2002, Clinical utility of sonography in the diagnosis and treatment of placental abruption, J. Ultrasound Med. source

      Not checked in the free run.

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

    Interpretationreasoning

    A timely diagnosis of placental abruption is vital for preventing adverse perinatal outcomes.

    “Therefore, a timely diagnosis of placental abruption is vital for preventing adverse perinatal outcomes.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    The kernelized hyperplane is adjusted by the SVM algorithms to provide maximum separation of data categories.

    “The kernelized hyperplane is adjusted by the SVM algorithms to provide maximum separation of data categories.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Data fusion in the anchor layers combines the data for further processing in the next layers.

    “The data fusion in the anchor layers combines the data for further processing in the next layers [23].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [23] Yang, 2019, Multilevel features convolutional neural network for multifocus image fusion, IEEE Trans. Comput. Imaging source

      Not checked in the free run.

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

    Premisecitation

    Multi-path CNN structures can fuse extracted features in multiple depths of layers.

    “Also, the multi-path CNN structures can fuse the extracted features in multiple depth of layers [22].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [22] Yuan, 2018, A multiscale and multidepth convolutional neural network for remote sensing imagery pan-sharpening, IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. source

      Not checked in the free run.

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

    Interpretationnone given

    The parallel bypassing nodes of the MP-ResNet-50 allow for preserving the highly weighted outputs of previous layers by passing them to the next layers directly.

    “The parallel bypassing nodes of this network allow for preserving the highly weighted outputs of previous layers by passing them to the next layers directly.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    The structure of the MP-ResNet-50 would keep the high-resolution details of the images by avoiding the convolutional filter and rectifying layers.

    “The structure of the MP-ResNet-50 would keep the high-resolution details of the images by avoiding the convolutional filter and rectifying layers.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    The guided filter is optimal for placental image processing because it often smooths away small structures not related to the placenta and eliminates noise and interferences common in ultrasound imaging.

    “The guided filter is optimal for placental image processing because it often smooths away the small structures that are not related to the placenta and eliminates the noise and interferences that are common in ultrasound imaging.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    The original data space is mapped into the kernel function space to overcome the non-linear separation of hyperplanes using high dimensional parameters.

    “The original data space is mapped into the kernel function k(x,y) space to overcome the non-linear separation of hyperplanes using high dimensional parameters.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    The higher effective resolution would be beneficial for the detection of anomalies and symptoms.

    “The higher effective resolution would be beneficial for the detection of anomalies and symptoms.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Eighty percent of the case and control sessions were randomly selected for training and the remaining 20% of the data were used for testing.

    “Eighty percent of the case and control sessions were randomly selected for training and the remaining 20% of the data were used for testing, respectively.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    This method can be utilized by clinicians to monitor the status of the placenta in real-time during routine prenatal clinical visits before the condition worsens.

    “This method can be utilized by clinicians to monitor the status of the placenta in real-time during routine prenatal clinical visits before the condition worsens.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    SVM and Ensemble classifiers are not efficient for analysis of two-dimensional inputs like images.

    “Since the SVM and Ensemble classifiers are not efficient for analysis of two-dimensional inputs like images”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Convolutional neural networks can directly process two-dimensional images.

    “However, convolutional neural networks can directly process two-dimensional images.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    The images were collected from various Kaiser Permanente hospitals and healthcare facilities with the same type of ultrasound machines.

    “The images are collected from various Kaiser Permanente hospitals and healthcare facilities with the same type of ultrasound machines.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    The multi-resolution structure of the parallel paths in the MP-CNN allows for the extraction of high-resolution and low-resolution information at the same time.

    “The multi-resolution structure of the parallel paths in this network allows for the extraction of high-resolution and low-resolution information at the same time.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    The covariance matrix in the guided filter extracts the fundamental area of the placental region.

    “The covariance matrix in the guided filter extracts the fundamental area of the placental region”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    Five-fold cross-validation reduces overfitting and improves the robustness of classifiers for training data with a high degree of diversity.

    “Five-fold cross-validation is used for all classifiers in this work which can result in the reduction of the overfitting as much as possible and in obtaining more information from the training feature vectors. This approach also improves the robustness of the classifiers for the training data with a high degree of diversity.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    The effect that the number of feature vector elements is optimized to improve performance is common for quantitative features.

    “It is shown that this effect is common for quantitative features [29].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [29] Khalvati, 2019, Prognostic value of CT radiomic features in resectable pancreatic ductal adenocarcinoma, Nat. Sci. Rep.

      Not checked in the free run.

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

    Interpretationreasoning

    The selection of the feature elements is related to the sequential combination of the components that include some stochastic measures.

    “Therefore, the selection of the feature elements is related to the sequential combination of the components that include some stochastic measures.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The components of the feature vector are not independent in general.

  • May combine two claims.

    Interpretationreasoning

    The reason for the session-level underperformance is that the variance of the input vectors applied to classifiers is reduced as the result of feature vector extraction; extraction of a single vector with the highest variance from all the feature vectors of the patient reduces the diversity of the inputs.

    “The reason for these results is that the variance of the input vectors that are applied to classifiers is reduced as the result of feature vector extraction. Extraction of a single vector with the highest variance from all the feature vectors of the patient reduces the diversity of the inputs.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Both the SVM and the decision tree ensemble are more dependent on the number of features and tuning of the parameters, while the CNN classifiers show less dependency.

    “Both the SVM and the decision tree ensemble are more dependent on the number of features and tunning of the parameters while the CNN classifiers show less dependency.”
    Quote in the paperSubstantial claim
33 of 33

  • The paper relies on this without support.

    Compared to CT, MRI, and PET, ultrasound has lower resolution, which makes analysis of ultrasound images more challenging.

    “Compared to other imaging approaches including Computerized Tomography (CT), Magnetic Resonance (MRI), and Positron Emission Tomography (PET), the lower resolution of ultrasound waves makes the analysis of ultrasound images more challenging.”

  • The paper relies on this without support.

    The components of the feature vector are not independent in general.

    “The components of the feature vector are not independent in general.”

  • The paper relies on this without support.

    The SVM and Ensemble classifiers are not efficient for analysis of two-dimensional inputs like images.

    “The SVM and Ensemble classifiers are not efficient for analysis of two-dimensional inputs like images.”

  • The paper relies on this without support.

    The identification of placental abruption through general ultrasonography images is a challenge.

    “The identification of the placental abruption through the general ultrasonography images is a challenge.”
4 of 4

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.

  • Discussion gives a different Inception-V3 training accuracy

    Major

    The discussion calls 87.40% the highest training accuracy, but the results report 87.68% for Inception-V3.

  • Image-level RFE feature counts disagree with the figure caption

    Minor

    The results text gives 52 features for SVM and 72 for Ensemble in the image-level analysis, while the Fig. 7 caption gives 55 and 52, respectively.

  • Introduction says previous studies were limited to MRI despite citing ultrasound studies

    Minor

    The introduction describes previous studies as limited to MRI, but its own review describes prior ultrasound-based placenta segmentation.

  • Selected-feature session results are described as full-length results

    Minor

    Table 2 assigns 88.87% AUC and 80.12% accuracy to the selected-feature session-level Ensemble model. Its all-features row instead reports 89.13% and 80.65%.

4 of 4

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. Huo, 2019, Stochastic tissue window normalization of deep learning on computed tomography, J. Med. Imaging, 6, 10.1117/1.JMI.6.4.044005 · DOI

Cited by

Sorted by citations

Related

Authors
Asadpour, V., Puttock, E. J., Getahun, D., Fassett, M. J., & Xie, F.
Affiliations
Kaiser Permanente (US)
Corresponding author
Asadpour, Vahid
ORCID
Venue
Heliyon
Publisher
Elsevier BV
Published
2023-02
Volume
9
Issue
2
Pages
e13577
ISSN
2405-8440 (ISSN-L)
DOI
10.1016/j.heliyon.2023.e13577
PMC
PMC9957707
OpenAlex
W4319868778
Cited by
11
References
23
Citations by year
2026: 2 · 2025: 3 · 2024: 4 · 2023: 2
FWCI
1.00
Type
journal article
Language
en
Open access
Licence
Locations
  • Heliyon (publishedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Kaiser Permanente
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Medical Image Segmentation Techniques
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
  • Health Sciences › Medicine › Radiology, Nuclear Medicine and Imaging › Radiomics and Machine Learning in Medical Imaging
Keywords
placental abruption, automated detection, prenatal ultrasound, semantic segmentation, placenta segmentation, support vector machine, ensemble classifiers, recursive feature elimination, ResNet50, InceptionV3, deep learning classification, pregnancy complications

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