arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新，休息一下吧，周末愉快~~
arXiv 2407.07557eess.IVcs.CV
收藏

基于知识蒸馏 Transformer 的真实世界心脏 CT 成像联邦学习

Real World Federated Learning with a Knowledge Distilled Transformer for Cardiac CT Imaging

DZHK (German Centre for Cardiovascular Research), all partner sites Department of Cardiology, Angiology；Pneumology, Heidelberg University Hospital, Heidelberg, Germany Informatics for Life Institute, Heidelberg, Germany Clinic for Cardiology III, University Hospital M\"unster, M\"unster, Germany Department of Medicine I, LMU University Hospital, LMU Munich, Munich, Germany Department of Diagnostic；Nuclear Medicine, University Medical Center Hamburg-Eppendorf, Hamburg, Germany Department of Medical Statistics, University Medical Center G\"ottingen, G\"ottingen, Germany Deutsches Herzzentrum der Charité (DHZC), Institute of Computer-assisted Cardiovascular Medicine, Berlin, Germany Charité – Universit\"atsmedizin Berlin, corporate member of Freie Universit\"at Berlin；Humboldt-Universit\"at zu Berlin, Berlin, Germany Fraunhofer Institute for Digital Medicine MEVIS, Bremen, Germany Institute of Bioinformatics, University Medicine Greifswald, Greifswald, Germany Institute for Experimental；Translational Cardiovascular Imaging, Goethe University, Frankfurt am Main, Germany Department of Cardiology, University Heart；Vascular Center Hamburg, University Medical Center Hamburg-Eppendorf, Hamburg, Germany Department of Cardiology, University Medicine G\"ottingen, G\"ottingen, Germany Department of Cardiology, Campus Kerckhoff of the Justus-Liebig-University at Gießen, Kerckhoff-Clinic, Gießen, Germany
Malte Tölle, Philipp Garthe, Clemens Scherer, Jan Moritz Seliger, Andreas Leha, Nina Krüger, Stefan Simm, Simon Martin, Sebastian Eble, Halvar Kelm, Moritz Bedn… 展开作者

Malte Tölle, Philipp Garthe, Clemens Scherer, Jan Moritz Seliger, Andreas Leha, Nina Krüger, Stefan Simm, Simon Martin, Sebastian Eble, Halvar Kelm, Moritz Bednorz, Florian André, Peter Bannas, Gerhard Diller, Norbert Frey, Stefan Groß, Anja Hennemuth, Lars Kaderali, Alexander Meyer, Eike Nagel, Stefan Orwat, Moritz Seiffert, Tim Friede, Tim Seidler, Sandy Engelhardt

阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出一种两步半监督知识蒸馏策略，将任务特定CNN的知识蒸馏至Transformer，在八家医院真实环境下进行最大规模联邦心脏CT分析，有效利用未标注数据提升预测准确性与泛化能力。

摘要：

联邦学习是一种在保护隐私的同时利用去中心化数据的著名技术。然而，现实应用常面临部分标注数据集的挑战，仅少数机构拥有特定专家标注，导致大量未标注数据未被使用。利用这些数据可增强 Transformer 架构在小型且标注多样的数据集环境中的能力。我们在八家医院的真实世界环境中进行了迄今为止最大规模的联邦心脏 CT 分析（n=8,104）。我们的两步半监督策略将任务特定 CNN 的知识蒸馏到 Transformer 中。首先，CNN 按标签类型对未标注数据进行预测，随后 Transformer 通过特定标签头从这些预测中学习。这提高了预测准确性，实现了联邦内所有部分标签的同步学习，并在下游任务的泛化能力上优于基于 UNet 的模型。代码和模型权重已开源以促进未来的心脏 CT 分析。

英文摘要：

Federated learning is a renowned technique for utilizing decentralized data while preserving privacy. However, real-world applications often face challenges like partially labeled datasets, where only a few locations have certain expert annotations, leaving large portions of unlabeled data unused. Leveraging these could enhance transformer architectures ability in regimes with small and diversely annotated sets. We conduct the largest federated cardiac CT analysis to date (n=8,104) in a real-world setting across eight hospitals. Our two-step semi-supervised strategy distills knowledge from task-specific CNNs into a transformer. First, CNNs predict on unlabeled data per label type and then the transformer learns from these predictions with label-specific heads. This improves predictive accuracy and enables simultaneous learning of all partial labels across the federation, and outperforms UNet-based models in generalizability on downstream tasks. Code and model weights are made openly available for leveraging future cardiac CT analysis.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。

↑