Beyond benchmarks of IUGC: Rethinking requirements of deep learning method for intrapartum ultrasound biometry from fetal ultrasound videos

A significant proportion (45%) of maternal deaths, neonatal deaths, and stillbirths occur during the intrapartum phase, particularly prevalent in low- and middle-income countries. Intrapartum biometry plays a crucial role in monitoring labor progress. However, the routine use of ultrasound in resource-limited settings is hindered by a shortage of trained sonographers. To tackle this issue, the Intrapartum Ultrasound Grand Challenge (IUGC), co-hosted with MICCAI 2024, was launched. The IUGC designed a multi-task automatic measurement framework oriented towards clinical applications. This framework integrates standard plane classification, fetal head-pubic symphysis segmentation, and biometry, enabling algorithms to leverage complementary information for more accurate estimations. Moreover, the challenge introduced the largest multi-center intrapartum ultrasound video dataset, consisting of 774 videos (68,106 images) collected from three hospitals. This rich dataset provides a solid foundation for algorithm training and evaluation. In this study, we elaborate on the details of the challenge, review the works submitted by eight teams, and interpret their methods from five aspects: preprocessing, data augmentation, learning strategy, model architecture, and post-processing. Additionally, we analyze the results considering various factors to identify key obstacles, explore potential solutions, and highlight ongoing challenges for future research. We conclude that although promising results have been achieved, the research remains in its early stages, and further in-depth exploration is required before clinical implementation. The solutions and the complete dataset are publicly accessible, aiming to drive continuous advancements in automatic biometry for intrapartum ultrasound imaging.