期刊
IEEE TRANSACTIONS ON MEDICAL IMAGING
卷 40, 期 3, 页码 1065-1077出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TMI.2020.3046692
关键词
Image segmentation; Training; Task analysis; Data models; Training data; Lesions; Sensitivity; Overfitting; class imbalance; image segmentation
类别
资金
- European Research Council (ERC) through the European Union's Horizon 2020 Research and Innovation Programme (Project MIRA, ERC-2017-STG) [757173]
- China Scholarship Council (CSC)
- U.K. Research and Innovation (UKRI) London Medical Imaging & Artificial Intelligence Centre for Value-Based Healthcare
Class imbalance in image segmentation neural networks can lead to overfitting, where the distribution of logit activations may shift across the decision boundary at test time, resulting in systematic under-segmentation of small structures. New asymmetric variants of popular loss functions and regularization techniques have been introduced to counteract this bias, demonstrating significantly improved segmentation accuracy compared to baselines and alternative approaches through extensive experiments.
Class imbalance poses a challenge for developing unbiased, accurate predictive models. In particular, in image segmentation neural networks may overfit to the foreground samples from small structures, which are often heavily under-represented in the training set, leading to poor generalization. In this study, we provide new insights on the problem of overfitting under class imbalance by inspecting the network behavior. We find empirically that when training with limited data and strong class imbalance, at test time the distribution of logit activations may shift across the decision boundary, while samples of the well-represented class seem unaffected. This bias leads to a systematic under-segmentation of small structures. This phenomenon is consistently observed for different databases, tasks and network architectures. To tackle this problem, we introduce new asymmetric variants of popular loss functions and regularization techniques including a large margin loss, focal loss, adversarial training, mixup and data augmentation, which are explicitly designed to counter logit shift of the under-represented classes. Extensive experiments are conducted on several challenging segmentation tasks. Our results demonstrate that the proposed modifications to the objective function can lead to significantly improved segmentation accuracy compared to baselines and alternative approaches.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据