Deep learning based medical image segmentation requires high quality pixel-level labeled data, which demands significant time and cost. Most existing semi-supervised learning methods exclude pseudo-labels with high uncertainty from training. This lead...
Deep learning based medical image segmentation requires high quality pixel-level labeled data, which demands significant time and cost. Most existing semi-supervised learning methods exclude pseudo-labels with high uncertainty from training. This leads to class imbalance issues and prevents learning in-depth representations beyond the class level. In this study, we propose a semi-supervised medical image segmentation method based on contrastive learning that leverages uncertainty information. The proposed method achieved DSC and Jaccard scores of 89.48 and 81.50, respectively, when using only 10% labeled data, outperforming existing methods.