超声视频中结节自动识别与区分以促进逐结节检查
图像与视频处理
2025-11-25 v2 机器学习
摘要
超声是健康筛查中的重要诊断技术，具有无创、经济、无辐射等优势，因此广泛应用于结节诊断。然而，它高度依赖超声医师的专业知识与临床经验。在超声图像中，单一结节在不同截面视图中可能呈现异质表现，难以进行逐结节检查。超声医师通常借助结节特征及腺体、导管等周围结构来区分不同结节，繁琐且耗时。为解决该问题，我们收集了数百段乳腺超声视频，构建了结节重识别系统，其包含两部分：基于深度学习模型、从输入视频片段提取特征向量的提取器，以及按结节自动聚类特征向量的实时聚类算法。系统取得满意结果并展现出区分超声视频的能力。据我们所知，这是重识别技术应用于超声领域的首次尝试。
引用
@article{arxiv.2310.06339,
title = {Automatic nodule identification and differentiation in ultrasound videos to facilitate per-nodule examination},
author = {Siyuan Jiang and Yan Ding and Yuling Wang and Lei Xu and Wenli Dai and Wanru Chang and Jianfeng Zhang and Jie Yu and Jianqiao Zhou and Chunquan Zhang and Ping Liang and Dexing Kong},
journal= {arXiv preprint arXiv:2310.06339},
year = {2025}
}
备注
The authors wish to withdraw this manuscript as it requires major revisions that substantially change the methodology and conclusions. A significantly updated version of this work may be submitted elsewhere at a later date. Thank you for your understanding