In this presentation, Lee Cooper describes a crowdsourcing approach in data collection for nucleus classification, localization and segmentation in breast cancer. He reviews a protocol that enables scalable generation of nuclear boundaries without laborious manual tracing and demonstrates that algorithmic suggestions improve the accuracy of non-expert annotations. A technique called Decision Tree Approximation of Learned Embeddings leverages nucleus segmentation data to improve the interpretability and the adoption of nucleus classification models is also outlined.
Get the presentation in seconds. Fill in the form opposite.