Prototype Contrastive Consistency Learning for Semi-Supervised Medical Image Segmentation

TL;DR

The paper addresses semi-supervised medical image segmentation under limited labeled data by introducing PCCS, a framework that fuses prototype-level contrastive learning with boundary-aware representations derived from a signed distance map. It advances three core ideas: (1) uncertainty-weighted prototype contrastive learning to tighten intra-class and separate inter-class prototypes, (2) a prototype updating mechanism that preserves prototype diversity via a Mean Teacher–inspired update with historical teacher information, and (3) an uncertainty-guided consistency loss to reduce prediction noise during pseudo-labeling. Together, these components yield state-of-the-art or competitive results across BUSI, BML, and ACDC datasets, including improved performance on boundary-rich malignant lesions and robust multi-class segmentation. The approach provides a practical, scalable path toward more accurate medical image segmentation with limited annotations, and the authors release their code for community use.

Abstract

Medical image segmentation is a crucial task in medical image analysis, but it can be very challenging especially when there are less labeled data but with large unlabeled data. Contrastive learning has proven to be effective for medical image segmentation in semi-supervised learning by constructing contrastive samples from partial pixels. However, although previous contrastive learning methods can mine semantic information from partial pixels within images, they ignore the whole context information of unlabeled images, which is very important to precise segmentation. In order to solve this problem, we propose a novel prototype contrastive learning method called Prototype Contrastive Consistency Segmentation (PCCS) for semi-supervised medical image segmentation. The core idea is to enforce the prototypes of the same semantic class to be closer and push the prototypes in different semantic classes far away from each other. Specifically, we construct a signed distance map and an uncertainty map from unlabeled images. The signed distance map is used to construct prototypes for contrastive learning, and then we estimate the prototype uncertainty from the uncertainty map as trade-off among prototypes. In order to obtain better prototypes, based on the student-teacher architecture, a new mechanism named prototype updating prototype is designed to assist in updating the prototypes for contrastive learning. In addition, we propose an uncertainty-consistency loss to mine more reliable information from unlabeled data. Extensive experiments on medical image segmentation demonstrate that PCCS achieves better segmentation performance than the state-of-the-art methods. The code is available at https://github.com/comphsh/PCCS.

Figures (10)

• Figure 1: Workflow of prototype contrastive learning. We aim to enhance the contrastive ability of prototypes to improve the effectiveness of contrastive learning.
• Figure 2: Overview of the PCCS. Arrows of different colors indicate the processing flow of different data. PCCS includes three modules, they are prototype contrastive learning module, prototype guide prototype module, and uncertainty-guided consistency learning module. The prototype contrastive learning module obtains contrastive samples from the feature map of the encoder and performs uncertainty-weighted prototype contrastive consistency loss $l_{pc}$. The prototype guide prototype module can enhance the diversity of prototype and improve generalization ability by aux loss $l_{aux}$. Uncertainty-guided consistency learning module enforces the model to make a consistent prediction for the output of two branches and reduce the uncertainty of the prediction by uncertainty-consistency loss $l_{c}$.
• Figure 3: Overview of prototype contrastive consistency loss. Large weights are assigned to positive prototypes with high similarity to the anchor, while small weights are assigned to positive prototypes with low similarity to the anchor. In this way, the positive samples with high similarity are kept close to the anchor and the positive samples with low similarity are suppressed, which can alleviate the impact of wrong pseudo-labels and learn a good feature space structure to achieve intra-class compactness and inter-class separability.
• Figure 4: Overview of the Prototype Guide Prototype Module.
• Figure 5: Visual comparison of segmentation results using different methods on the BUSI dataset.