ACTION++: Improving Semi-supervised Medical Image Segmentation with Adaptive Anatomical Contrast

Chenyu You,Weicheng Dai,Yifei Min,Lawrence Staib,Jasjeet S. Sekhon,James S. Duncan

2023-07-18

Abstract:Medical data often exhibits long-tail distributions with heavy class imbalance, which naturally leads to difficulty in classifying the minority classes (i.e., boundary regions or rare objects). Recent work has significantly improved semi-supervised medical image segmentation in long-tailed scenarios by equipping them with unsupervised contrastive criteria. However, it remains unclear how well they will perform in the labeled portion of data where class distribution is also highly imbalanced. In this work, we present ACTION++, an improved contrastive learning framework with adaptive anatomical contrast for semi-supervised medical segmentation. Specifically, we propose an adaptive supervised contrastive loss, where we first compute the optimal locations of class centers uniformly distributed on the embedding space (i.e., off-line), and then perform online contrastive matching training by encouraging different class features to adaptively match these distinct and uniformly distributed class centers. Moreover, we argue that blindly adopting a constant temperature $\tau$ in the contrastive loss on long-tailed medical data is not optimal, and propose to use a dynamic $\tau$ via a simple cosine schedule to yield better separation between majority and minority classes. Empirically, we evaluate ACTION++ on ACDC and LA benchmarks and show that it achieves state-of-the-art across two semi-supervised settings. Theoretically, we analyze the performance of adaptive anatomical contrast and confirm its superiority in label efficiency.

Computer Vision and Pattern Recognition,Artificial Intelligence,Machine Learning,Image and Video Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the class - imbalance problem in medical image segmentation due to the long - tailed nature of data distribution, especially the difficulty in classifying the minority classes (such as boundary regions or rare objects). Although existing semi - supervised learning methods have made significant progress in dealing with unlabeled data, it is still unclear how they perform when there is also a highly unbalanced class distribution in the labeled data part. Therefore, the paper proposes an improved contrastive learning framework ACTION++, which improves the effect of semi - supervised medical image segmentation through adaptive anatomical contrast, especially when dealing with long - tailed distribution datasets. Specifically, ACTION++introduces an adaptive supervised contrastive loss and proposes a strategy of using a dynamic temperature parameter to improve the separation between the majority and minority classes, thereby achieving better segmentation results.