Dual-Criteria Active Learning for Medical Image Segmentation

In recent years, deep learning-based medical image segmentation models have achieved significant success in various tasks. However, their training typically relies on a large amount of labeled medical data, which often poses a challenge due to the high labeling cost in practical applications. To address this issue, active learning methods select the most informative samples from a large pool of unlabeled medical data for annotation, thereby reducing the amount of labeled data required. Studies have shown that models trained using these highly informative labeled samples often perform similarly to models trained on the entire dataset. This paper proposes a dual-criterion active learning method, which efficiently combines uncertainty and representativeness as evaluation metrics to effectively select the most informative samples, thereby enhancing the training efficiency and performance of medical image segmentation models.