Deep learning for digital and virtual histology

点击翻译按钮获取中文摘要

The proposed project will leverage the cutting-edge advances in tissue imaging techniques and machine learning to develop next generation virtual and digital pathology toolsets with artificial intelligence-based algorithms: (a) label-free virtual histological imaging and (2) time-efficient cancerous tissue intraoperative diagnosis system on the envisioned virtual images. It includes a partnership between University of Alberta researchers Roger Zemp and Xingyu Li and a Machine Learning company, AltaML. Particularly, it will leverage novel imaging technology pioneered by the Zemp group to form virtual histological images of tissues without cutting, fixing, or staining procedures. The Li group will develop deep-learning based stain-transfer models to render such virtual histological images so that they are virtually indistinguishable from true gold-standard pathology images and investigate deep intraoperative diagnostic models that incorporate explainable artificial intelligence (XAI) to determine the presence and locations of tumor tissues in margins of resected tissues. In the long-term, the goal is to provide a surgeon with a tool to assess whether there may be residual tumor tissues which need removal, using this novel imaging technology along with deep learning toolsets. Along this journey, a secondary goal is to develop deep learning methods that can provide automated histological decision making and generate heatmaps to direct a pathologist's attention for rapid review and accelerated workflow.