One-Step Abductive Multi-Target Learning with Diverse Noisy Samples: An Application to Tumour Segmentation for Breast Cancer

One-step abductive multi-target learning (OSAMTL) is an approach proposed to handle complex noisy labels. However, OSAMTL is not suitable for the situation where diverse noisy samples (DNS) are provided for a learning task. In this paper, giving definition of DNS, we propose one-step abductive multi-target learning with DNS (OSAMTL-DNS) to expand the original OSAMTL to a wider range of tasks that handle complex noisy labels. Applying OSAMTL-DNS to tumour segmentation for breast cancer in medical histopathology whole slide image analysis, we show that OSAMTL-DNS is able to enable various state-of-the-art approaches for learning from noisy labels to achieve significantly more rational predictions.