Learning to Segment from Noisy Annotations: A Spatial Correction Approach
Learning to Segment from Noisy Annotations: A Spatial Correction Approach
复制标题
DOI：
10.48550/arxiv.2308.02498
复制
发表时间：
2023-07
期刊：
影响因子：
--
通讯作者：
Jiacheng Yao;Yikai Zhang;Songzhu Zheng;Mayank Goswami;P. Prasanna;Chao Chen
中科院分区：
文献类型：
--
作者：
Jiacheng Yao;Yikai Zhang;Songzhu Zheng;Mayank Goswami;P. Prasanna;Chao Chen
噪声标签会显著影响深度神经网络的性能。在医学图像分割任务中，由于标注时间和对标注者专业知识的要求较高，标注容易出错。现有方法大多假设不同像素的噪声标签为\textit{I.I.D}。然而，分割标签噪声通常具有很强的空间相关性，在分布上有明显的偏倚。在本文中，我们提出了一种新的马尔可夫模型分割噪声注释，编码空间相关和偏差。为了消除标签噪声，我们提出了一种逐步恢复真实标签的标签校正方法。我们为所提出的方法的正确性提供了理论保证。实验表明，我们的方法在合成和现实世界的噪声注释上都优于当前最先进的方法。
Noisy labels can significantly affect the performance of deep neural networks (DNNs). In medical image segmentation tasks, annotations are error-prone due to the high demand in annotation time and in the annotators' expertise. Existing methods mostly assume noisy labels in different pixels are \textit{i.i.d}. However, segmentation label noise usually has strong spatial correlation and has prominent bias in distribution. In this paper, we propose a novel Markov model for segmentation noisy annotations that encodes both spatial correlation and bias. Further, to mitigate such label noise, we propose a label correction method to recover true label progressively. We provide theoretical guarantees of the correctness of the proposed method. Experiments show that our approach outperforms current state-of-the-art methods on both synthetic and real-world noisy annotations.