OSAL-ND: Open-Set Active Learning for Nucleus Detection

The recent advance of deep learning has shown promising power for nucleus detection that plays an important role in histopathological examination. However, such accurate and reliable deep learning models need enough labeled data for training, which makes active learning (AL) an attractive learning paradigm for reducing the annotation efforts by pathologists. In open-set environments, AL encounters the challenge that the unlabeled data usually contains non-target samples from the unknown classes, resulting in the failure of most AL methods. Although AL has been explored in many open-set classification tasks, research on AL for nucleus detection in the open-set environment remains unexplored. To address the above issues, we propose a two-stage AL framework designed for nucleus detection in an open-set environment (i.e., OSAL-ND). In the first stage, we propose a prototype-based query strategy based on the auxiliary detector to select a candidate set from known classes as pure as possible. In the second stage, we further query the most uncertain samples from the candidate set for the nucleus detection task relying on the target detector. We evaluate the performance of our method on the NuCLS dataset, and the experimental results indicate that our method can not only improve the selection quality on the known classes, but also achieve higher detection accuracy with lower annotation burden in comparison with the existing studies.