2608.16855
2026-08-18
cs.CV
新提交
89%
Can Unsupervised Methods Outperform Supervised Deep Learning When Ground Truth Is Sparse? A Case Study of Bronchovascular Bundle Segmentation in Low-Dose CT
当真实标注稀疏时,无监督方法能否优于监督深度学习?——基于低剂量CT中支气管血管束分割的案例研究
Anna Mrukwa (1), Marek Socha (1), Aleksandra Suwalska (1), Agata Durawa (2), Malgorzata Jelitto (3), Katarzyna Dziadziuszko (3), Edyta Szurowska (3), Pawel Bozek (4), Michal Marczyk (1 and 5), Witold Rzyman (2), Rafal Dziadziuszko (6), Joanna Polanska (1) ((1) Department of Data Science and Engineering, Silesian University of Technology, Gliwice, Poland, (2) Department of Thoracic Surgery, Medical University of Gdansk, Gdansk, Poland, (3) 2nd Division of Radiology, Medical University of Gdansk, Gdansk, Poland, (4) Department of Radiology and Radiodiagnostics, Medical University of Silesia, Katowice, Poland, (5) Department of Breast Medical Oncology, Yale School of Medicine, New Haven, CT, USA, (6) Department of Oncology and Radiotherapy, Medical University of Gdansk, Gdansk, Poland)
专题命中
医学影像
:CT(title,title_cn);分类 cs.CV
AI总结
本研究以低剂量CT的支气管血管束分割为案例,构建RONALD无监督分割流程,在真实标注稀疏时,其结节保留率显著优于现有方法,可提升极早期肺癌的肺结节检测效果。