From the Institute and Polyclinic for Diagnostic and Interventional Radiology (L.M.), Else Kröner Fresenius Center for Digital Health (L.M., J.N.K.), and Department of Medicine I (J.N.K.), Faculty of Medicine and University Hospital Carl Gustav Carus, TUD Dresden University of Technology, Fetscherstrasse 74, 01307 Dresden, Germany; Department of Diagnostic and Interventional Radiology, University Hospital Aachen, Aachen, Germany (G.M.F., D.T.); and Medical Oncology, National Center for Tumor Diseases (NCT), University Hospital Heidelberg, Heidelberg, Germany (J.N.K.).
Radiology. 2024 Jul;312(1):e232085. doi: 10.1148/radiol.232085.
Deep learning (DL) is currently the standard artificial intelligence tool for computer-based image analysis in radiology. Traditionally, DL models have been trained with strongly supervised learning methods. These methods depend on reference standard labels, typically applied manually by experts. In contrast, weakly supervised learning is more scalable. Weak supervision comprises situations in which only a portion of the data are labeled (incomplete supervision), labels refer to a whole region or case as opposed to a precisely delineated image region (inexact supervision), or labels contain errors (inaccurate supervision). In many applications, weak labels are sufficient to train useful models. Thus, weakly supervised learning can unlock a large amount of otherwise unusable data for training DL models. One example of this is using large language models to automatically extract weak labels from free-text radiology reports. Here, we outline the key concepts in weakly supervised learning and provide an overview of applications in radiologic image analysis. With more fundamental and clinical translational work, weakly supervised learning could facilitate the uptake of DL in radiology and research workflows by enabling large-scale image analysis and advancing the development of new DL-based biomarkers.
深度学习（DL）目前是放射学中基于计算机的图像分析的标准人工智能工具。传统上，DL 模型是使用强监督学习方法进行训练的。这些方法依赖于参考标准标签，通常由专家手动应用。相比之下，弱监督学习更具可扩展性。弱监督包括以下情况：只有一部分数据被标记（不完全监督），标签是指整个区域或病例，而不是精确划定的图像区域（不精确监督），或者标签包含错误（不准确监督）。在许多应用中，弱标签足以训练有用的模型。因此，弱监督学习可以为训练 DL 模型解锁大量原本无法使用的数据。一个例子是使用大型语言模型从放射学报告的自由文本中自动提取弱标签。在这里，我们概述了弱监督学习的关键概念，并概述了其在放射图像分析中的应用。随着更基础和临床转化工作的开展，弱监督学习可以通过实现大规模图像分析和推进新的基于 DL 的生物标志物的开发，促进 DL 在放射学和研究工作流程中的应用。
