CoChat

DilatedFormer: dilated granularity transformer network for placental maturity grading in ultrasound

Open access

The DOI matches its Crossref record; 47 of the 60 references it cites were found in a registry or an index; 26 of the 44 sources its claims cite were checked for support.

Abstract

Placental maturity grading is often utilized for evaluating fetal growth and maternal health. The paper presents a dilated granularity transformer network and verifies it on 500 ultrasound images.

Overview

CoChat Summary

TL;DR

The paper introduces DilatedFormer, an end-to-end transformer-based network for placental maturity grading (PMG) in ultrasound that uses dilated self-attention, a detail-specific feed-forward layer, and a feature pyramid decoder. On a dataset of 500 annotated ultrasound images, it reports better performance than compared methods, reaching 0.89 accuracy, 0.9574 Cohen’s kappa, and 0.9022 F1-score.

Key findings

  • The collected dataset contains 500 ultrasound images: 128 Grade I, 115 Grade II, 122 Grade III, and 135 Grade IV.
  • On the test setup described, DilatedFormer achieved 0.89 accuracy, 0.9574 Cohen’s kappa, and 0.9022 F1-score, outperforming all compared methods.
  • Compared with the best-performing baselines among the listed methods, the paper reports improvement from 0.85 to 0.89 in accuracy, from 0.9407 to 0.9574 in Cohen’s kappa, and from 0.8921 to 0.9022 in F1-score.
  • Ablation results show the full model (“Basic + DSA + FPD + Detail-specific”) improved over the baseline (“Basic”) from 0.85 to 0.89 accuracy, 0.9365 to 0.9574 kappa, and 0.8904 to 0.9022 F1-score.

Methods

  • The model is a four-stage transformer architecture with dilated transformer blocks using multi-head dilated self-attention, a detail-specific feed-forward network, and a feature pyramid decoder for PMG prediction.
  • The study uses a collected and annotated ultrasound dataset of 500 images, taken from the anterior wall placenta using GE Voluson E8 Expert/Phillip EPIQ7 systems; images were split into 400 training and 100 test images.
  • Training used PyTorch, Adam, 80 epochs, initial learning rate 1×10^-4, batch size 16, input size 224×224, and evaluation used accuracy, Cohen’s kappa, and F1-score.

Limitations

  • The paper states that the model is trained on ultrasound images with the same intensity distribution in a fully supervised fashion, and performance may degrade at new clinical sites due to vendor variability.
  • The authors state future work is needed to make the model more lightweight and improve computational complexity.
  • The text also suggests a limitation in data efficiency/generalization, since the authors say they should develop a more generalizable algorithm and incorporate multi-modality information such as clinical data.

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

Integrity signals

Registry

Crossref · 10.3389/fphy.2023.1239400

Cites

47 of 47 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

14 Valid · 1 Not valid · 7 Not checked · 24 Own results and interpretations

17 more claims · 2 with their cited sources checked

Create a free account to see all 46 claims and to verify the 2 references still open.

See all claims — free account
  • It may state two separate claims as one. 8
  • Probably not a substantial claim. 17
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Citations

Cites

Cited by

See all 2

Metadata

Authors
Wu, Y., Yang, Y., Zhu, L., Han, Z., Luo, H., Xue, X., & Wang, W.
Affiliations
  • Sichuan University (CN)
  • The Hong Kong University of Science and Technology (Guangzhou) (CN)
  • Third Affiliated Hospital of Sun Yat-sen University (CN)
  • Jiangxi University of Water Resources and Electric Power (CN)
  • Hong Kong Metropolitan University (HK)
Corresponding author
Luo, Hong
Venue
Frontiers in Physics
Published
2023-09-04
Volume
11
ISSN
2296-424X (ISSN-L)
OpenAlex
W4386417425
Cited by
2
References
60
Citations by year
2026: 1 · 2025: 1
FWCI
0.15
Type
journal article
Language
en
Open access
Locations
  • Frontiers in Physics (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Rare & Special e-Zone (The Hong Kong University of Science and Technology) (submittedVersion)
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 › Health Informatics › Artificial Intelligence in Healthcare and Education
Keywords
ultrasound images, end-to-end learning, fetal growth assessment, medical image classification

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

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

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.