Keywords: Fully Convolutional Neural Network, Weak Annotation, Instance Segmentation
Abstract: The applicability of Deep Learning based methods to image segmentation tasks in general, and to nuclei segmentation in particular, is currently limited by the effort required to collect large enough high-quality training data. In this work we describe a novel deep-learning based approach for instance segmentation that utilizes an easier to collect weak training data. Namely, a mixture of fully segmented nuclei and nuclei with only center locations specified. We demonstrate the robustness of our proposed approach - using 30% of fully segmented nuclei decreases the algorithm performance by only 4%.
Code Of Conduct: I have read and accept the code of conduct.
Remove If Rejected: Remove submission from public view if paper is rejected.