CoChat

Medical Professional Enhancement Using Explainable Artificial Intelligence in Fetal Cardiac Ultrasound Screening

Open access

The DOI matches its Crossref record; 73 of the 74 references it cites were found in a registry or an index.

Abstract

Diagnostic support tools based on artificial intelligence (AI) have exhibited high performance in various medical fields. However, their clinical application remains challenging because of the lack of explanatory power in AI decisions (black box problem), making it difficult to build trust with medical professionals. Nevertheless, visualizing the internal representation of deep neural networks will increase explanatory power and improve the confidence of medical professionals in AI decisions. We propose a novel deep learning-based explainable representation “graph chart diagram” to support fetal cardiac ultrasound screening, which has low detection rates of congenital heart diseases due to the difficulty in mastering the technique. Screening performance improves using this representation from 0.966 to 0.975 for experts, 0.829 to 0.890 for fellows, and 0.616 to 0.748 for residents in the arithmetic mean of area under the curve of a receiver operating characteristic curve. This is the first demonstration wherein examiners used deep learning-based explainable representation to improve the performance of fetal cardiac ultrasound screening, highlighting the potential of explainable AI to augment examiner capabilities.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.3390/biomedicines10030551

Cites

73 of 73 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure A1 Process to compute a graph chart diagram with a simple auto-encoder. Kernel K slides over the barcode-like timeline in the time direction (red open arrow) and slices the detection status of the required substructure. The substructure detection status is input to the auto-encoder and compressed into two dimensions.
Figure A1 Process to compute a graph chart diagram with a simple auto-encoder. Kernel K slides over the barcode-like timeline in the time direction (red open arrow) and slices the detection status of the required substructure. The substructure detection status is input to the auto-encoder and compressed into two dimensions.CC BY · PubMed Central
Figure 1 Flow chart of the proposed method. The screening videos were scanned in one direction from the stomach to the head. Therefore, a video contains diagnostic planes from the abdomen to the four-chamber view (4CV) and three-vessel trachea view (3VTV). The barcode-like timeline represents the detection of 18 substructures Rall for each frame of the screening video arranged in the time direction. The kernel slices the barcode-like timeline and feeds it to the cascade graph encoder. The kernel moves in the time direction (red open arrow). A sub-graph chart diagram was created for each of the four vessels Rvessels and the eight-heart substructures Rheart. Information on the six other substructures Rothers ( =Rall\(Rvessels∪Rheart) is appended to two sub-graph chart diagrams to obtain the main graph chart diagrams. The view-proxy loss is applied to the main-graph chart diagram so that the main graph chart diagram is generated stably. The gray circle dots in the main-graph chart correspond to the parts of the barcode-like timeline that was sliced into the kernel at a certain time. In particular, the green dot indicates the point corresponding to the green kernel and moves in the direction of the red arrow. An abnormality score is calculated using the main-graph chart diagram and provided to the examiner. Rall is the set of { crux, ventricular septum, right atrium, tricuspid valve, right ventricle, left atrium, mitral valve, left ventricle, pulmonary artery, ascending aorta, superior vena cava, descending aorta, stomach, spine, umbilical vein, inferior vena cava, pulmonary vein, ductus arteriosus}. Rvessels is the set of the {pulmonary artery, ascending aorta, superior vena cava, ductus arteriosus} Rheart is the set of {crux, ventricular septum, right atrium, tricuspid valve, right ventricle, left atrium, mitral valve, left ventricle}. Rothers set of Rall\(Rvessels∪Rheart).
Figure 1 Flow chart of the proposed method. The screening videos were scanned in one direction from the stomach to the head. Therefore, a video contains diagnostic planes from the abdomen to the four-chamber view (4CV) and three-vessel trachea view (3VTV). The barcode-like timeline represents the detection of 18 substructures Rall for each frame of the screening video arranged in the time direction. The kernel slices the barcode-like timeline and feeds it to the cascade graph encoder. The kernel moves in the time direction (red open arrow). A sub-graph chart diagram was created for each of the four vessels Rvessels and the eight-heart substructures Rheart. Information on the six other substructures Rothers ( =Rall\(Rvessels∪Rheart) is appended to two sub-graph chart diagrams to obtain the main graph chart diagrams. The view-proxy loss is applied to the main-graph chart diagram so that the main graph chart diagram is generated stably. The gray circle dots in the main-graph chart correspond to the parts of the barcode-like timeline that was sliced into the kernel at a certain time. In particular, the green dot indicates the point corresponding to the green kernel and moves in the direction of the red arrow. An abnormality score is calculated using the main-graph chart diagram and provided to the examiner. Rall is the set of { crux, ventricular septum, right atrium, tricuspid valve, right ventricle, left atrium, mitral valve, left ventricle, pulmonary artery, ascending aorta, superior vena cava, descending aorta, stomach, spine, umbilical vein, inferior vena cava, pulmonary vein, ductus arteriosus}. Rvessels is the set of the {pulmonary artery, ascending aorta, superior vena cava, ductus arteriosus} Rheart is the set of {crux, ventricular septum, right atrium, tricuspid valve, right ventricle, left atrium, mitral valve, left ventricle}. Rothers set of Rall\(Rvessels∪Rheart).CC BY · PubMed Central
Figure 2 Schematic diagram of the effect of view-proxy loss. (a) Graph chart diagram without view-proxy loss. The graph chart diagram (grey connected circles) without view-proxy loss is formed, in which points corresponding to actual 4CV (surrounded by the blue line), actual 3VTV (surrounded by the red line), and other planes are randomly mixed. (b) Graph chart diagram with view-proxy loss. The stars denote the coordinates for the ideal 4CV (blue) and 3VTV (red). The squares denote corresponding points to the synthesized barcode-like timelines for 4CV (blue) and 3VTV (red). The view-proxy loss (orange bidirectional arrows) considers the loss between the stars and squares. As a corollary of optimizing the view-proxy loss, the grey points surrounded by the blue or red lines are attracted to their respective specified coordinates. As a result of optimizing the view-proxy loss, the graph chart diagram is expected to be untangled.
Figure 2 Schematic diagram of the effect of view-proxy loss. (a) Graph chart diagram without view-proxy loss. The graph chart diagram (grey connected circles) without view-proxy loss is formed, in which points corresponding to actual 4CV (surrounded by the blue line), actual 3VTV (surrounded by the red line), and other planes are randomly mixed. (b) Graph chart diagram with view-proxy loss. The stars denote the coordinates for the ideal 4CV (blue) and 3VTV (red). The squares denote corresponding points to the synthesized barcode-like timelines for 4CV (blue) and 3VTV (red). The view-proxy loss (orange bidirectional arrows) considers the loss between the stars and squares. As a corollary of optimizing the view-proxy loss, the grey points surrounded by the blue or red lines are attracted to their respective specified coordinates. As a result of optimizing the view-proxy loss, the graph chart diagram is expected to be untangled.CC BY · PubMed Central
Figure 3 Main-graph chart diagrams. The main-graph chart diagrams were obtained from the fetal cardiac ultrasound screening video of a normal case (a) and tetralogy of Fallot case (b). The dots circled in gray correspond to each kernel, and these dots are connected as the kernel moves. Several points correspond to the typical diagnostic planes. The four-chamber view (4CV) and three-vessel trachea view (3VTV) are shown as blue diamond and red square points, respectively.
Figure 3 Main-graph chart diagrams. The main-graph chart diagrams were obtained from the fetal cardiac ultrasound screening video of a normal case (a) and tetralogy of Fallot case (b). The dots circled in gray correspond to each kernel, and these dots are connected as the kernel moves. Several points correspond to the typical diagnostic planes. The four-chamber view (4CV) and three-vessel trachea view (3VTV) are shown as blue diamond and red square points, respectively.CC BY · PubMed Central
Figure 4 Sub-graph chart diagrams. (a,b) show sub-graph chart diagrams of the vessels and heart in the normal video, respectively. (d,e) show the sub-graph chart diagrams of the vessels and heart in the TOF video, respectively. (c,f) are the same as (a,b) in Figure 3, and are presented for reference. The blue diamond points correspond to the 4CV, and the red square points correspond to the 3VTV. TOF, tetralogy of Fallot; 4CV, four-chamber view; 3VTV, three-vessel trachea view.
Figure 4 Sub-graph chart diagrams. (a,b) show sub-graph chart diagrams of the vessels and heart in the normal video, respectively. (d,e) show the sub-graph chart diagrams of the vessels and heart in the TOF video, respectively. (c,f) are the same as (a,b) in Figure 3, and are presented for reference. The blue diamond points correspond to the 4CV, and the red square points correspond to the 3VTV. TOF, tetralogy of Fallot; 4CV, four-chamber view; 3VTV, three-vessel trachea view.CC BY · PubMed Central
Figure 5 Performance of the AI-assisted fetal cardiac ultrasound screening. The ROC curves show the screening performance comparison of experts (a), fellows (b), and residents (c) with and without AI assistance. The color of the solid lines represents the ROC curve for AI only (gray), examiner only (green), and examiner with AI (red). Solid lines indicate mean values. Lighter semitransparent colors surrounding the the solid lines indicate the standard deviation. The mean and standard deviation of the AUC of the ROC curves are also reported in the legends. ROC, receiver operating characteristics; AUC, area under the curve; TPR, true positive rate; FPR, false-positive rate.
Figure 5 Performance of the AI-assisted fetal cardiac ultrasound screening. The ROC curves show the screening performance comparison of experts (a), fellows (b), and residents (c) with and without AI assistance. The color of the solid lines represents the ROC curve for AI only (gray), examiner only (green), and examiner with AI (red). Solid lines indicate mean values. Lighter semitransparent colors surrounding the the solid lines indicate the standard deviation. The mean and standard deviation of the AUC of the ROC curves are also reported in the legends. ROC, receiver operating characteristics; AUC, area under the curve; TPR, true positive rate; FPR, false-positive rate.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

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.

The self-check needs the paper’s full text.

Citations

Cites

Show all 73 references

Research works printed but not linked: 2

  1. Arnaout, 2021, An ensemble of neural networks provides expert-level prenatal detection of complex congenital heart disease, Nat. Med., 27, 882, 10.1038/s41591-021-01342-5 · doi:10.1038/s41591-021-01342-5 · Paper page
  2. Goodfellow, I., Bengio, Y., and Courville, A. (2016). Deep Learning, MIT Press.

Cited by

See all 68

Metadata

Authors
Sakai, A., Komatsu, M., Komatsu, R., Matsuoka, R., Yasutomi, S., Dozen, A., Shozu, K., Arakaki, T., Machino, H., Asada, K., Kaneko, S., Sekizawa, A., & Hamamoto, R.
Affiliations
  • Tokyo Medical and Dental University (JP)
  • RIKEN Center for Advanced Intelligence Project (JP)
  • SHOWA Medical University (JP)
  • Fujitsu (Japan) (JP)
  • National Cancer Center Research Institute (JP)
Corresponding author
  • Komatsu, Masaaki
  • Hamamoto, Ryuji
Venue
Biomedicines
Publisher
MDPI AG
Published
2022-02-25
Volume
10
Issue
3
Pages
551
ISSN
2227-9059 (ISSN-L)
OpenAlex
W4214574368
Cited by
68
References
74
Citations by year
2026: 8 · 2025: 15 · 2024: 17 · 2023: 23 · 2022: 5
FWCI
4.94
Type
journal article
Language
en
Open access
Locations
  • Biomedicines (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
Ministry of Education, Culture, Sports, Science and Technology (Advanced Integrated Intelligence Platform)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Health Sciences › Health Professions › Health Information Management › Artificial Intelligence in Healthcare
Keywords
explainable AI, fetal echocardiography, deep learning, black box problem, diagnostic support tool, medical training

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.

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Medical Professional Enhancement Using Explainable Artificial Intelligence in Fetal Cardiac Ultrasound Screening. https://app.cochat.ai/papers/10.3390/biomedicines10030551
The paper
Sakai, A., Komatsu, M., Komatsu, R., Matsuoka, R., Yasutomi, S., Dozen, A., Shozu, K., Arakaki, T., Machino, H., Asada, K., Kaneko, S., Sekizawa, A., & Hamamoto, R. (2022). Medical Professional Enhancement Using Explainable Artificial Intelligence in Fetal Cardiac Ultrasound Screening. Biomedicines. https://doi.org/10.3390/biomedicines10030551

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.