Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FUGC: Benchmarking Semi-Supervised Learning Methods for Cervical Segmentation

Jieyun Bai, Yitong Tang, Zihao Zhou, Mahdi Islam, Musarrat Tabassum, Enrique Almar-Munoz, Hongyu Liu, Hui Meng, Nianjiang Lv, Bo Deng, Yu Chen, Zilun Peng, Yusong Xiao, Li Xiao, Nam-Khanh Tran, Dac-Phu Phan-Le, Hai-Dang Nguyen, Xiao Liu, Jiale Hu, Mingxu Huang, Jitao Liang, Chaolu Feng, Xuezhi Zhang, Lyuyang Tong, Bo Du, Ha-Hieu Pham, Thanh-Huy Nguyen, Min Xu, Juntao Jiang, Jiangning Zhang, Yong Liu, Md. Kamrul Hasan, Jie Gan, Zhuonan Liang, Weidong Cai, Yuxin Huang, Gongning Luo, Mohammad Yaqub, Karim Lekadir

TL;DR

The paper presents FUGC, the first benchmark for semi-supervised cervical segmentation in transvaginal ultrasound, addressing critical data-scarcity issues in PTB risk assessment. It provides a standardized 890-image dataset with a clear evaluation protocol (DSC, HD, RT) and a weighted ranking scheme that prioritizes accuracy while maintaining practical runtimes. Across 10 top-performing semi-supervised methods, the study demonstrates that well-designed pseudo-label refinement, consistency training, and ensemble strategies yield high segmentation accuracy (DSC_All around 0.90–0.93) with fast inference times (as low as tens to hundreds of milliseconds). The benchmark promotes reproducibility, offers valuable insights into method design for ultrasound, and supports the clinical translation of AI-assisted cervical length assessment for PTB risk prediction.

Abstract

Accurate segmentation of cervical structures in transvaginal ultrasound (TVS) is critical for assessing the risk of spontaneous preterm birth (PTB), yet the scarcity of labeled data limits the performance of supervised learning approaches. This paper introduces the Fetal Ultrasound Grand Challenge (FUGC), the first benchmark for semi-supervised learning in cervical segmentation, hosted at ISBI 2025. FUGC provides a dataset of 890 TVS images, including 500 training images, 90 validation images, and 300 test images. Methods were evaluated using the Dice Similarity Coefficient (DSC), Hausdorff Distance (HD), and runtime (RT), with a weighted combination of 0.4/0.4/0.2. The challenge attracted 10 teams with 82 participants submitting innovative solutions. The best-performing methods for each individual metric achieved 90.26\% mDSC, 38.88 mHD, and 32.85 ms RT, respectively. FUGC establishes a standardized benchmark for cervical segmentation, demonstrates the efficacy of semi-supervised methods with limited labeled data, and provides a foundation for AI-assisted clinical PTB risk assessment.

FUGC: Benchmarking Semi-Supervised Learning Methods for Cervical Segmentation

TL;DR

The paper presents FUGC, the first benchmark for semi-supervised cervical segmentation in transvaginal ultrasound, addressing critical data-scarcity issues in PTB risk assessment. It provides a standardized 890-image dataset with a clear evaluation protocol (DSC, HD, RT) and a weighted ranking scheme that prioritizes accuracy while maintaining practical runtimes. Across 10 top-performing semi-supervised methods, the study demonstrates that well-designed pseudo-label refinement, consistency training, and ensemble strategies yield high segmentation accuracy (DSC_All around 0.90–0.93) with fast inference times (as low as tens to hundreds of milliseconds). The benchmark promotes reproducibility, offers valuable insights into method design for ultrasound, and supports the clinical translation of AI-assisted cervical length assessment for PTB risk prediction.

Abstract

Accurate segmentation of cervical structures in transvaginal ultrasound (TVS) is critical for assessing the risk of spontaneous preterm birth (PTB), yet the scarcity of labeled data limits the performance of supervised learning approaches. This paper introduces the Fetal Ultrasound Grand Challenge (FUGC), the first benchmark for semi-supervised learning in cervical segmentation, hosted at ISBI 2025. FUGC provides a dataset of 890 TVS images, including 500 training images, 90 validation images, and 300 test images. Methods were evaluated using the Dice Similarity Coefficient (DSC), Hausdorff Distance (HD), and runtime (RT), with a weighted combination of 0.4/0.4/0.2. The challenge attracted 10 teams with 82 participants submitting innovative solutions. The best-performing methods for each individual metric achieved 90.26\% mDSC, 38.88 mHD, and 32.85 ms RT, respectively. FUGC establishes a standardized benchmark for cervical segmentation, demonstrates the efficacy of semi-supervised methods with limited labeled data, and provides a foundation for AI-assisted clinical PTB risk assessment.
Paper Structure (43 sections, 5 figures, 7 tables)

This paper contains 43 sections, 5 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Challenges
  3. Motivation
  4. Related Work
  5. Related Grand Challenges
  6. State-of-the-art Methods for Cervical Segmentation
  7. Materials and Methods
  8. Dataset
  9. Evaluation Metrics and Ranking Scheme
  10. Metrics
  11. Ranking Scheme
  12. Challenge Entries
  13. Top Ten Solutions
  14. T1
  15. T2
  16. ...and 28 more sections

Figures (5)

  • Figure 1: The anterior and posterior lips are key anatomical structures for identifying the cervix, despite their wide variability in size and shape.
  • Figure 2: The FUGC workflow, showing the segmentation focus, challenge timeline, participants, and dataset.
  • Figure 3: Dice Similarity Coefficients for the anterior lip (DSC_A), posterior lip (DSC_P), and the overall cervix (DSC_All). (Ai-Ci) Dot- and boxplots for visualizing the evaluation metric data. (Aii-Cii) Significance maps for visualizing the results of significance testing. (Aiii-Ciii) Blob plots for visualizing ranking stability. (Aiv-Civ) Line plots for visualizing rankings robustness across different ranking methods.
  • Figure 4: Hausdorff Distances for the anterior lip (HD_A), posterior lip (HD_P), and the overall cervix (HD_All). (Ai-Ci) Dot- and boxplots for visualizing the evaluation metric data. (Aii-Cii) Significance maps for visualizing the results of significance testing. (Aiii-Ciii) Blob plots for visualizing ranking stability. (Aiv-Civ) Line plots for visualizing rankings robustness across different ranking methods.
  • Figure 5: (A) Top-model region map, showing the winning team for each $(w_{\text{mDSC}}, w_{\text{mHD}})$ pair, with $w_{\text{RT}} = 1 - w_{\text{mDSC}} - w_{\text{mHD}}$.(B) Pareto frontier illustrating trade-offs among mDSC, mHD, and RT scores.