Learning from certain regions of interest in medical images via probabilistic positive-unlabeled networks

The laborious annotation process and inherent image ambiguity exacerbate difficulties of data acquisition for medical image segmentation, leading to suboptimal performance in practice. This paper proposes a workaround against these challenges aiming to learn unbiased models solely from certainties. Concretely, during the labeling stage, only regions of interest confidently discerned by annotators are required to be labeled, not only increasing label quantity but also improving label quality. This approach formulates the positive-unlabeled (PU) segmentation problem and motivates to capture uncertainty in ambiguous regions. We thus delve into data-generating assumptions in the PU segmentation context and propose Probabilistic PU Segmentation Networks (ProPU-Nets) to tackle problems abovementioned. This framework employs the expectation-maximization algorithm to gradually estimate true masks, and more importantly, by encoding plausible segmentation variants in a latent space, uncertainty estimation can be naturally embedded into the PU segmentation framework. Benefitting from the framework's unbiasedness, a semi-supervised PU segmentation method is also proposed, which can further excavate performance gains from unlabeled data. We conduct extensive experiments on LIDC, RIGA, and LA datasets, and comprehensively compared with state-of-the-art methods in label-efficient medical image segmentation. The results justify the method's effectiveness and practical prospect.