USPNet: a novel network for intrapartum ultrasound standard plane classification
计算机科学
超声波
医学
放射科
作者
Yifan Zhang,Jieyun Bai,Yaosheng Lu
出处
期刊:日期:2025-05-16卷期号:: 16-16被引量:1
标识
DOI:10.1117/12.3062734
摘要
This paper presents USPNet, a novel model designed for the classification of standard planes in intrapartum ultrasound images, focusing on improving the classification accuracy of small anatomical targets. Intrapartum ultrasound, as a non-invasive, objective, and reproducible method, is increasingly utilized to evaluate labor progress. However, due to the small size of key anatomical features, accurate classification remains challenging. USPNet addresses this challenge by introducing a simplified architecture that incorporates a Shift block, SC-MLP, and ORIConv blocks to enhance feature extraction. Experimental results on two datasets, Zj-AOP and Zj-CD, demonstrate that USPNet outperforms existing advanced models in terms of accuracy, precision, recall, and F1-score. The proposed model effectively improves the classification of small anatomical targets, providing enhanced support for clinical decision-making during labor. Code are available at https://github.com/2458zhanghehe/USPNet.