SSL-FetalBioNet: Self-Supervised Learning for Automated Angle of Progression Measurement in Intrapartum Ultrasound

WangKun

Keywords: Self-supervised Learning· U-Net · Intrapartum Ultrasound · Angle of Progress

Abstract:

During childbirth, real-time assessment of fetal head position and progression is crucial for ensuring the safety of both mother and
infant. Detecting key anatomical landmarks in intrapartum ultrasound
images and calculating the Angle of progression (AoP) have become critical techniques in the next-generation childbirth monitoring protocol proposed by the World Health Organization (WHO). However, traditional
manual analysis is time-consuming and prone to subjective bias, highlighting the urgent need for automated methods to achieve standardized
and precise childbirth assessment. This paper presents a key point detection approach combining self-supervised pre-training with a U-Net
architecture: first, the encoder is pre-trained using large-scale unlabeled
images through self-supervision to uncover latent structural information;
subsequently, this pre-trained encoder is transferred to the supervised
learning stage to achieve precise localization of three key points (PS1,
PS2, FH1). Our method achieved eighth place in the Intrapartum Ultrasound Grand Challenge 2025, demonstrating its effectiveness and generalization capability in the task of key point detection in intrapartum
ultrasound. This work provides a practical and feasible pathway toward
automated and scalable childbirth monitoring, with significant implications for global maternal and infant health.

Submission Number: 9