Nuclear Segmentation and Classification: On Color and Compression Generalization

Since the introduction of digital and computational pathology as a field, one major problems in clinical application algorithms has been struggle to generalize well examples outside distribution training data. Existing work address this both natural images focused almost exclusively on classification tasks. We explore evaluate robustness 7 best performing nuclear segmentation models from largest challenge for problem date, CoNIC challenge. demonstrate that existing state-of-the-art (SoTA) are robust towards compression artifacts but suffer substantial performance reduction when subjected shifts color domain. find using stain normalization domain shift can be detrimental model performance. On other hand, neural style transfer is more consistent improving test presented with large variations wild.