CoChat
Sign Up
Cited by 63Congenital Heart Disease StudiesArtificial Intelligence in Healthcare and EducationArtificial Intelligence in Healthcare

Deep Learning for Improving the Effectiveness of Routine Prenatal Screening for Major Congenital Heart Diseases

Open access
Nurmaini, Siti, Partan, Radiyati Umi, Bernolian, Nuswil, Sapitri, Ade Iriani, Tutuko, Bambang, Rachmatullah, Muhammad Naufal, Darmawahyuni, Annisa, Firdaus, Firdaus +1 more·2022·Journal of Clinical Medicine
DOI 10.3390/jcm11216454PMC PMC9653675
Open this paper in CoChat — free accountOpen PDF

Abstract

Early prenatal screening with an ultrasound (US) can significantly lower newborn mortality caused by congenital heart diseases (CHDs). However, the need for expertise in fetal cardiologists and the high volume of screening cases limit the practically achievable detection rates. Hence, automated prenatal screening to support clinicians is desirable. This paper presents and analyses potential deep learning (DL) techniques to diagnose CHDs in fetal USs. Four convolutional neural network architectures were compared to select the best classifier with satisfactory results. Hence, dense convolutional network (DenseNet) 201 architecture was selected for the classification of seven CHDs, such as ventricular septal defect, atrial septal defect, atrioventricular septal defect, Ebstein’s anomaly, tetralogy of Fallot, transposition of great arteries, hypoplastic left heart syndrome, and a normal control. The sensitivity, specificity, and accuracy of the DenseNet201 model were 100%, 100%, and 100%, respectively, for the intra-patient scenario and 99%, 97%, and 98%, respectively, for the inter-patient scenario. We used the intra-patient DL prediction model to validate our proposed model against the prediction results of three expert fetal cardiologists. The proposed model produces a satisfactory result, which means that our model can support expert fetal cardiologists to interpret the decision to improve CHD diagnostics. This work represents a step toward the goal of assisting front-line sonographers with CHD diagnoses at the population level.

AI snapshot

Written from the paper’s full text.

TL;DR

This paper compares four CNN architectures for classifying seven major congenital heart diseases plus normal controls from fetal four-chamber ultrasound views, selecting DenseNet201 as the best model. It reports high intra-patient performance and compares the model against three expert fetal cardiologists, while noting inter-patient limitations and the need for multi-center validation.

Key findings

  • DenseNet201 was selected from DenseNet121, DenseNet201, ResNet50, and ResNet101; for intra-patient classification it achieved 100% sensitivity, 100% specificity, and 100% accuracy, and for inter-patient classification the abstract reports 99% sensitivity, 97% specificity, and 98% accuracy.
  • In the eight-class task, intra-patient DenseNet201 classified all classes, but in the inter-patient scenario TOF was misclassified as ASD, EA, and TGA, and VSD was misclassified as TOF and TGA; after augmentation, TOF and VSD were classified successfully and classifier performance increased by 30–35%.
  • Against three expert fetal cardiologists, agreement with the CNN model was Kappa 0.912 for the first expert, 0.542 for the second, and 0.669 for the third, with the first expert’s interpretation described as almost equal to the proposed CNN prediction.
  • Data augmentation increased training data from 1129 to 23,504 echocardiograms (19,626 for training and the rest for validation), and Section 3.4 reports inter-patient DenseNet201 performance of 92% accuracy, 91% sensitivity, and 92% specificity versus RLDS at 91% accuracy and 91% sensitivity.

Methods

  • The study included 76 pregnant women at Mohammad Hoesin General Hospital, Palembang, Indonesia: 31 case patients and 45 normal controls; fetal heart images were collected prospectively by different operators using GE Voluson E6 machines as DICOM videos, with frames resized from 640 × 480 to 300 × 300 pixels.
  • Data were divided into intra-patient and inter-patient scenarios: about 1129 echocardiograms for training/testing in the intra-patient scenario and about 55 echocardiograms for testing in the inter-patient scenario; the expert fetal cardiologist’s interpretation served as the gold standard.
  • Four CNN architectures were trained and tested without augmentation initially, using learning rate 0.001, SoftMax activation, Adam optimization, categorical_crossentropy loss, dropout 0.5, Python/PyTorch 1.7.1; evaluation used confusion-matrix metrics including accuracy, sensitivity, specificity, PPV, and NPV, with Grad-CAM plus guided backpropagation for explainability.

Limitations

  • The paper states that the limited number of echocardiograms made the learning process on DenseNet201 produce an insignificant result, and that TOF and VSD were difficult to distinguish because of similar structural abnormalities.
  • The paper states that expanded model testing in multiple centers and more training data from more centers will be necessary in the future.
  • Evident from the text: the dataset comes from a single hospital with a small subject sample, and the reported inter-patient metrics differ between the abstract (99% sensitivity, 97% specificity, 98% accuracy) and Section 3.4/Table 8 (92% accuracy, 91% sensitivity, 92% specificity).

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

Integrity signals

Registry

Crossref · 10.3390/jcm11216454

Cites

21 of 21 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 20 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Data augmentation increased the classifier performance by 30-35%, especially in the inter-patient scenario.

    “Based on such a process, the classifier performance is increased by 30–35%, especially in the inter-patient scenario (Table 6).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.
    • Rests on an assumption: Augmented echocardiograms remain clinically valid representations of CHD.
  • Findingown data

    In the inter-patient scenario, two classes are unclassified.

    “However, in the inter-patient scenario, two classes are unclassified (Figure 5b).”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.
  • Findingreasoning

    From validation with inter-patient data using four architectures, none of them succeeded in recognizing TOF and VSD.

    “From the validation with inter-patient data using such four architectures, none of them succeeded in recognizing TOF and VSD”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.
  • Findingown data

    In intra-patient scenarios, the whole classes can be classified by the DenseNet201 model.

    “In intra-patient scenarios, the whole classes can be classified by the DenseNet201 model (Figure 5a).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.
  • Findingown data

    The classifier model achieved the best performance in diagnosing fetal CHDs in intra-patient data, with approximately 100% accuracy, sensitivity, and specificity.

    “Table 8 illustrates that our classifier model achieved the best performance in diagnosing fetal CHDs in intra-patient data, with approximately 100% accuracy, sensitivity, and specificity.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.
  • May combine two claims. 20
  • Classified as probably not a substantial claim. 9

  • May combine two claims.

    Findingown data

    The class TOF is misclassified as ASD, EA, and TGA, and the class VSD is misclassified as TOF and TGA.

    “The class TOF is misclassified as ASD, EA, and TGA, and the class VSD is misclassified as TOF and TGA.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

  • May combine two claims.

    Findingown data

    After data augmentation, TOF and VSD are classified successfully, whereas they previously failed to classify.

    “In addition, TOF and VSD are classified successfully that previously failed to classify (Figure 5b and Figure 6).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.
    • Rests on an assumption: Augmented echocardiograms remain clinically valid representations of CHD.

  • May combine two claims.

    Findingown data

    DenseNet201 results outperformed those of state-of-the-art methods in most comparisons and provided classification of CHDs versus normal.

    “However, DenseNet201 results outperformed those of state-of-the-art methods in most of them, and the results provided a classification of CHDs versus normal.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

  • May combine two claims.

    Findingown data

    DenseNet201's performance on eight-class classification outperforms other architectures and two-class classification.

    “It can be seen that DenseNet201’s performance on eight-class classification outperforms other architectures and two-class classification.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

  • May combine two claims.

    Findingown data

    The DenseNet201 model is significantly better than all four CNN architectures, with the highest performance among all state-of-the-art approaches.

    “Our DenseNet201 model is significantly better than all four CNN architectures, with the highest performance among all state-of-the-art approaches.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

  • May combine two claims.

    Findingown data

    In the intra-patient scenario, DenseNet201 achieved 100% sensitivity (85.75-100%), 100% specificity (93.62-100%), 100% positive predictive value, 100% negative predictive value, and 100% accuracy (95.49-100%).

    “However, the DenseNet201 reach had 100% sensitivity (85.75–100%), 100% specificity (93.62–100%), 100% positive predictive value, 100% negative predictive value, and 100% accuracy (95.49–100%; Table 3 and Table 4).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

  • May combine two claims.

    Findingown data

    The confusion matrix shows a perfect match between the predictions of the proposed model and the actual image, with no error prediction in both cases and controls.

    “The confusion matrix (CM) shows a perfect match between the predictions of the proposed model and the actual image, which means that there was no error prediction in both the cases or controls.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

  • May combine two claims.

    Findingown data

    When tested with inter-patient data, the performance of the proposed model was maintained, and the sensitivity, specificity, and accuracy values remained above 90%.

    “When tested with inter-patient data, the performance of the proposed model was maintained, and the sensitivity, specificity, and accuracy values remained above 90%.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

  • May combine two claims.

    Findingown data

    The proposed DenseNet201 classifier had 92% accuracy, 91% sensitivity, and 92% specificity on interpatient data, whereas the RLDS achieved 91% accuracy and 91% sensitivity.

    “Our proposed DenseNet201 classifier had 92% accuracy, 91% sensitivity, and 92% specificity, whereas the RLDS achieved 91% accuracy and 91% sensitivity.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

  • May combine two claims.

    Findingown data

    The second expert fetal cardiologist's Kappa value was about 0.542, with 99.73% positive predictive value and 62.23% negative predictive value.

    “However, the second and third expert fetal cardiologists only reached the Kappa value of about 0.542 (99.73% positive predictive value and 62.23% negative predictive value) and 0.669 (97.52% positive predictive value and 70.31% negative predictive value), respectively.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

  • May combine two claims.

    Findingown data

    The Kappa value between the first expert fetal cardiologist's interpretation and the proposed CNN predictions was about 0.912, with 99.73% positive predictive value and 91.74% negative predictive value.

    “The results of the Kappa test between the first expert fetal cardiologists’ interpretation and the proposed CNNs predictions demonstrated a substantial degree of conformity with the Kappa value of about 0.912, with a 99.73% positive predictive value and a 91.74% negative predictive value.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

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

    Findingnone given

    There was an excellent localization process for most of the images used in the experiment.

    “There was an excellent localization process for most of the images used in our experiment.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The third expert fetal cardiologist's Kappa value was about 0.669, with 97.52% positive predictive value and 70.31% negative predictive value.

    “However, the second and third expert fetal cardiologists only reached the Kappa value of about 0.542 (99.73% positive predictive value and 62.23% negative predictive value) and 0.669 (97.52% positive predictive value and 70.31% negative predictive value), respectively.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

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

    Premisecitation

    The top seven congenital heart diseases include ventricular septal defect, atrial septal defect, atrioventricular septal defect, Ebstein's anomaly, tetralogy of Fallot, transposition of great arteries, and hypoplastic left heart syndrome.

    “The top seven CHDs include ventricular septal defect (VSD), atrial septal defect (ASD), atrioventricular septal defect (AVSD), Ebstein’s anomaly (EA), tetralogy of Fallot (TOF), transposition of great arteries (TGA), and hypoplastic left heart syndrome (HLHS) [3,4].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [3] Puri, 2017, Congenital Heart Disease, Pediatr. Rev. source

      Not checked in the free run.

    • cites [4] Backer, 2016, Shunt Lesions Part I: Patent Ductus Arteriosus, Atrial Septal Defect, Ventricular Septal Defect, and Atrioventricular Septal Defect, Pediatr. Crit. Care Med. source

      Not checked in the free run.

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

    Study design and conductreasoning

    The learning process was conducted without augmentation data to maintain the actual clinical condition.

    “We conducted the learning process without augmentation data to maintain the actual clinical condition.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    The expert fetal cardiologist's interpretation of the entire ultrasound served as the gold standard for the learning process.

    “The expert fetal cardiologist’s interpretation of the entire US served as our gold standard for the learning process.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Expert fetal cardiologists can detect CHDs with over 98% sensitivity and near 90% specificity.

    “Expert fetal cardiologists can detect CHDs with over 98% sensitivity and near 90% specificity [9].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [9] Arnaout, 2021, An ensemble of neural networks provides expert-level prenatal detection of complex congenital heart disease, Nat. Med. source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Although some methods achieved good performance using the intra-patient scheme, their results are unreliable because their evaluation process was biased.

    “Therefore, although some of these methods achieved good performance using the intra-patient scheme, their results are unreliable because their evaluation process was biased [17].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [17] Liu, 2019, A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: A systematic review and meta-analysis, Lancet Digit. Health source

      Not checked in the free run.

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

    Interpretationnone given

    The limited number of echocardiograms makes the learning process on DenseNet 201 produce an insignificant result.

    “In addition, the limited number of echocardiograms makes the learning process on DenseNet 201 produce an insignificant result.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

  • May combine two claims.

    Premisecitation

    CHD screening can be started at 11 weeks of gestation using a 4-chamber view, and sensitivity can increase to 92% at 13 weeks of gestation.

    “CHD screening can be started at 11 weeks of gestation by using a 4-chamber view (4CV), and sensitivity can increase to 92% at 13 weeks of gestation [5,6].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [5] Norton, M.E. (2016). Callen’s Ultrasonography in Obstetrics and Gynecology, Elsevier Health Sciences.

      Not checked in the free run.

    • cites [6] Donofrio, 2014, Diagnosis and treatment of fetal cardiac disease: A scientific statement from the American Heart Association, Circulation source

      Not checked in the free run.

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

    Study design and conductreasoning

    Dropout (0.5) was used as a regularization technique to reduce overfitting.

    “For reducing overfitting, dropout (0.5) is used as a regularization technique.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Most previously reported works in the literature had been evaluated based on the intra-patient paradigm rather than the inter-patient scheme, which is a more realistic scenario to prevent training and testing the model using samples from the same patients.

    “Furthermore, most previously reported works in the literature had been evaluated based on the intra-patient paradigm rather than the inter-patient scheme, which is a more realistic scenario to prevent training and test the model using samples from the same patients [21].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [21] Crispi, 2020, Machine Learning in Fetal Cardiology: What to Expect, Fetal Diagn. Ther. source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    One type of abnormality in the TOF class is the complicated VSD, so the pixels on the TOF echocardiogram will be identical to the VSD class.

    “One type of abnormality in the TOF class is the complicated VSD, thus the pixels on the TOF echocardiogram will be identical to the VSD class.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

  • May combine two claims.

    Interpretationown data

    The proposed model successfully predicts CHDs with good performance, validated with inter-patient data and three expert fetal cardiologists.

    “The proposed model successfully predicts CHDs with good performance, validated with inter-patient data and three expert fetal cardiologists.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

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

    Interpretationreasoning

    Due to the small fetal heart size, the dynamics of the fetal position and the fetal heart take up a relatively small proportion of the image, meaning that the DL algorithm must learn to ignore a large proportion of the available data.

    “Due to the small fetal heart size, the dynamics of the fetal position and the fetal heart take up a relatively small proportion of the image, meaning that the DL algorithm must learn to ignore a large proportion of the available data.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Class-discriminative visualization enables expert fetal cardiologists to understand where models are predicted.

    “Class-discriminative visualization enables expert fetal cardiologists to understand where models are predicted.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The proposed model displays relatively higher accuracy in intra- and inter-patient data and is thus credible enough for the early diagnosis of fetal CHDs.

    “Correspondingly, our proposed model displays relatively higher accuracy in intra- and inter-patient data and is thus credible enough for the early diagnosis of fetal CHDs.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

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

    Premisecitation

    The RLDS architecture has three classes of CHDs (HRHS, HLHS, and highly RAS) versus the normal control.

    “Unfortunately, the RLDS architecture has three classes of CHDs (HRHS, HLHS, and highly RAS) versus the normal control.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [13] Qiao, 2022, RLDS: An explainable residual learning diagnosis system for fetal congenital heart disease, Futur. Gener. Comput. Syst. source

      Not checked in the free run.

28 of 28

  • The paper relies on this without support.

    The expert fetal cardiologist’s interpretation of the entire ultrasound is a valid gold standard for CHD diagnosis.

    “The expert fetal cardiologist’s interpretation of the entire US served as our gold standard for the learning process.”

  • The paper relies on this without support.

    No research has performed an automatic fetal heart diagnostic process for many CHD conditions.

    “No research has performed an automatic fetal heart diagnostic process for many CHDs conditions.”

  • The paper relies on this without support.

    Most CHD prediction methods use only one or two cases and one control.

    “Most CHD prediction methods use only one or two cases and one control.”

  • The paper relies on this without support.

    Early prenatal screening with ultrasound can significantly lower newborn mortality caused by congenital heart diseases.

    “Early prenatal screening with an ultrasound (US) can significantly lower newborn mortality caused by congenital heart diseases (CHDs).”

  • The paper relies on this without support.

    Augmented echocardiograms remain clinically valid representations of CHD.

    “The augmentation process is implemented by using echocardiogram geometric transformation. We serve image rotation (±15 degrees), height and width shift, and vertical and horizontal flipping.”
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?

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.

  • Gestational-history controls exceed the control-group size

    Major

    The gestational-history breakdown states that 46 controls had gravida 1-4, already more than the reported total of 45 controls.

  • Inter-patient DenseNet201 performance is reported differently

    Major

    The abstract reports inter-patient sensitivity, specificity, and accuracy of 99%, 97%, and 98%, while the later comparison gives 91%, 92%, and 92% for the proposed DenseNet201 classifier. No distinction between these inter-patient results is stated.

  • Parity categories exceed the control-group size

    Minor

    The three stated control parity categories total 65 subjects, rather than the reported 45 controls.

    Reported 45 · recomputed 65

3 of 3

Cites

Flagged first, then by citations

Research works printed but not linked: 3

  1. Norton, M.E. (2016). Callen’s Ultrasonography in Obstetrics and Gynecology, Elsevier Health Sciences.
  2. Ammirato, P., and Berg, A.C. (2019). A Mask-RCNN Baseline for Probabilistic Object Detection. arXiv.
  3. Arnaout, R., Curran, L., Zhao, Y., Levine, J.C., Chinn, E., and Moon-Grady, A.J. (2020). Expert-level prenatal detection of complex congenital heart disease from screening ultrasound using deep learning. medRxiv. · DOI · Paper page

Cited by

Sorted by citations

Related

Authors
Nurmaini, S., Partan, R. U., Bernolian, N., Sapitri, A. I., Tutuko, B., Rachmatullah, M. N., Darmawahyuni, A., Firdaus, F., & Mose, J. C.
Affiliations
  • Sriwijaya University (ID)
  • Internal Medicine, Mohammad Hoesin General Hospital, Palembang 30126, Indonesia
  • Division of Fetomaternal, Department of Obstetrics and Gynaecology, Mohammad Hoesin General Hospital, Palembang 30126, Indonesia
  • Padjadjaran University (ID)
Corresponding author
Nurmaini, Siti
ORCID
Venue
Journal of Clinical Medicine
Publisher
MDPI AG
Published
2022-10-31
Volume
11
Issue
21
Pages
6454
ISSN
2077-0383 (ISSN-L)
DOI
10.3390/jcm11216454
PMC
PMC9653675
OpenAlex
W4307989357
Semantic Scholar
3f60ebe04f239a1470ff3ac4d8cf53a00f288a27
Cited by
61
References
21
Citations by year
2026: 18 · 2025: 21 · 2024: 10 · 2023: 12
FWCI
7.76
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of Clinical Medicine (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Universitas Sriwijaya (Proffesional Grants 2022)
Topics
  • Health Sciences › Medicine › Epidemiology › Congenital Heart Disease Studies
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
Keywords
congenital heart disease, prenatal screening, prenatal ultrasound, deep learning, DenseNet201, convolutional neural networks, automated diagnosis, ventricular septal defect, atrial septal defect, tetralogy of Fallot, transposition of the great arteries, Ebstein's anomaly, hypoplastic left heart syndrome, atrioventricular septal defect, clinical decision support

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