Active Learning with the nnUNet and Sample Selection with Uncertainty-Aware Submodular Mutual Information Measure

Annotating medical images for segmentation tasks is a time-consuming process that requiresexpert knowledge. Active learning can reduce this annotation cost and achieve optimalmodel performance by selecting only the most informative samples for annotation. However, the eectiveness of active learning sample selection strategies depends on the modelarchitecture and training procedure used. The nnUNet has achieved impressive results invarious automated medical image segmentation tasks due to its self-configuring pipelinefor automated model design and training. This raises the question of whether the nnUNetis applicable in an active learning setting to avoid cumbersome manual configuration ofthe training process and improve accessibility for non-experts in deep learning-based segmentation. This paper compares various sample selection strategies in an active learningsetting in which the self-configuring nnUNet is used as the segmentation model. Additionally, we propose a new sample selection strategy for UNet-like architectures: USIM - Uncertainty-Aware Submodular Mutual Information Measure. The method combinesuncertainty and submodular mutual information to select batches of uncertain, diverse,and representative samples. We evaluate the performance gain and labeling costs on threemedical image segmentation tasks with different segmentation challenges. Our findingsdemonstrate that utilizing nnUNet as the segmentation model in an active learning setting is feasible, and most sampling strategies outperform random sampling. Furthermore,we demonstrate that our proposed method yields a significant improvement compared toexisting baseline methods.