Automated Annotator Variability Inspection for Biomedical Image Segmentation

Supervised deep learning approaches for automated diagnosis support require datasets annotated by experts. Intra-annotator variability of a single annotator and inter-annotator between annotators can affect the quality support. As medical experts will always differ in annotation details, quantitative studies concerning are particular interest. A consistent noise-free large-scale by, example, dermatologists or pathologists is current challenge. Hence, methods needed to automatically inspect annotations datasets. In this paper, we categorize noise image segmentation tasks, present simulate noise, examine impact on quality. Two novel identify intra-annotator inconsistencies based uncertainty-aware neural networks proposed. We demonstrate benefits our inspection such as focused re-inspection noisy detection generally different styles using biomedical ISIC 2017 Melanoma dataset.