[论文解读] Beyond Benchmarks of IUGC: Rethinking Requirements of Deep Learning Methods for Intrapartum Ultrasound Biometry from Fetal Ultrasound Videos

Jieyun Bai, Zihao Zhou+61 more|arXiv (Cornell University)|Feb 13, 2026

Fetal and Pediatric Neurological Disorders被引用 1

一句话总结

本文分析 Intrapartum Ultrasound Grand Challenge (IUGC) 2024，综述多任务深度学习在产时超声视频中的标准截面分类、FH-PS 分割与生物测量等任务的应用、数据集、方法与挑战。

ABSTRACT

A substantial proportion (45\%) of maternal deaths, neonatal deaths, and stillbirths occur during the intrapartum phase, with a particularly high burden in low- and middle-income countries. Intrapartum biometry plays a critical role in monitoring labor progression; however, the routine use of ultrasound in resource-limited settings is hindered by a shortage of trained sonographers. To address this challenge, the Intrapartum Ultrasound Grand Challenge (IUGC), co-hosted with MICCAI 2024, was launched. The IUGC introduces a clinically oriented multi-task automatic measurement framework that integrates standard plane classification, fetal head-pubic symphysis segmentation, and biometry, enabling algorithms to exploit complementary task information for more accurate estimation. Furthermore, the challenge releases the largest multi-center intrapartum ultrasound video dataset to date, comprising 774 videos (68,106 frames) collected from three hospitals, providing a robust foundation for model training and evaluation. In this study, we present a comprehensive overview of the challenge design, review the submissions from eight participating teams, and analyze their methods from five perspectives: preprocessing, data augmentation, learning strategy, model architecture, and post-processing. In addition, we perform a systematic analysis of the benchmark results to identify key bottlenecks, explore potential solutions, and highlight open challenges for future research. Although encouraging performance has been achieved, our findings indicate that the field remains at an early stage, and further in-depth investigation is required before large-scale clinical deployment. All benchmark solutions and the complete dataset have been publicly released to facilitate reproducible research and promote continued advances in automatic intrapartum ultrasound biometry.

研究动机与目标

• Motivate automated, end-to-end measurement of intrapartum ultrasound parameters to support labor management and reduce adverse outcomes.
• Provide a standardized multi-center intrapartum video benchmark for tasks including standard plane classification, FH-PS segmentation, and biometry (AoP and HSD).
• Review submitted methods across preprocessing, data augmentation, learning strategy, architecture, and post-processing to identify key bottlenecks and opportunities for improvement.
• Facilitate open scientific exchange by releasing datasets and code to drive future research in intrapartum ultrasound biometry.

提出的方法

• Describe a multitask framework that integrates standard plane classification, FH-PS segmentation, and biometry from ultrasound videos.
• Analyze challenge submissions and organize evaluation using the Biomedical Image Analysis ChallengeS (BIAS) methodology.
• Leverage a large multi-center video dataset (774 videos, 68,106 images) for training, validation, and testing.
• Provide public access to solutions and datasets to enable reproducibility and further development.
• Discuss use of template/landmark-based measurement for AoP and HSD within segmented FH and PS regions.
• Highlight challenges such as intra-class variation, low inter-class differences, ultrasound noise, and anatomical deformation during labor.

Fig. 1: Challenges faced by each task in the Intrapartum Ultrasound Grand Challenge (IUGC). Standard plane classification is affected by large intra-class variability caused by imaging artifacts, soft tissue deformation, fetal posture changes, and probe motion, as well as low inter-class separabilit

实验结果

研究问题

• RQ1 What are effective deep learning strategies for multi-task intrapartum ultrasound analysis (plane classification, FH-PS segmentation, and biometry) from video data?
• RQ2 How does a large multi-center intrapartum video dataset influence generalization and benchmark reliability for automatic biometry?
• RQ3 What are the main obstacles and potential solutions to achieving clinically usable automatic intrapartum biometry?
• RQ4 To what extent can public datasets and open-source baselines accelerate progress in end-to-end intrapartum ultrasound measurement?

主要发现

• The IUGC 2024 dataset comprises 774 videos (68,106 images) from three hospitals, with training/validation/test splits of 434/40/300 videos respectively.
• Eight teams submitted across classification, segmentation, and biometry with seven top-performing methods analyzed alongside a baseline.
• Multitask approaches were evaluated in a real-world, multi-center setting, highlighting the importance of integrating standard plane recognition with FH-PS segmentation for accurate biometry.
• Despite promising results, the study emphasizes that intrapartum ultrasound biometry remains in early stages and further work is required before clinical deployment.
• The authors provide public access to the complete dataset and the top-performing methods to foster continued advancement in automatic intrapartum biometry.
• The paper discusses ongoing clinical and technical challenges, including data heterogeneity, spatio-temporal modeling of ultrasound videos, and error propagation across cascaded tasks.

Fig. 2: Overall Workflow of Clinical Image Utilization and the Intrapartum Ultrasound Grand Challenge (IUGC). A) Clinical images are acquired via a transperineal ultrasound (US) approach using mid-sagittal scans from pregnant women during labor. B) Manual operations encompass: classification of stan