U-Net with SSM-Inspired Gated Large-Receptive-Field Refinement for Fetal Abdomen Segmentation and Circumference Measurement Using Ultrasound Images

Abstract

Ultrasound images are widely used to monitor fetal growth, but automatic segmentation of fetal structures is difficult due to noise, motion artifacts, and image variability. This study addresses these challenges using deep neural networks. This study contributes a U-shaped encoder-decoder model with state-space model (SSM)-inspired large-receptive-field refinement blocks for fetal abdomen segmentation and fetal abdomen circumference measurement. Experiments are conducted using the benchmark dataset, ACOUSLIC-AI [1], and by following five-fold cross validation and ten-fold cross validation strategies. Standard measures of Dice coefficient were used to evaluate performance in segmentation and the accuracy of the estimated abdominal circumference was evaluated using the Normalized Absolute Error(NAE-AC)-. © 2026 IEEE.