Automated Histopathologic Tissue Annotation with Deep Learning
Recent advances in Machine Learning and Computer Vision, in conjunction with the advent of whole-slide imaging, are fuelling the digital revolution in the centuries-old field of pathology. In a collaborative project between Roche's pRED Informatics and the department of Pathology and Tissue Analysis, we developed a fully integrated approach for the automated histopathologic segmentation of tissue samples in order to support increasing complexity and demand for tissue analysis in drug development. In a proof-of-concept study, we leveraged the extensive data generated on Roche’s IRIS platform and trained and deployed fully convolutional deep neural networks for pixel-level segmentation of immunohistochemically stained tissue sections from biopsies and surgical specimen. The evaluation of automated annotation of imaged tissue into healthy, cancerous and necrotic areas suggests performance on par with expert annotations.
Dr. Fabian Schmich works as a Data Scientist at the Roche Innovation Center Munich. His research is focused on Machine Learning, data mining and applied statistics in the context of cancer immunotherapy. Fabian studied computer science, mathematics and biology in Munich, Toronto, and London and received his diploma degree in bioinformatics from Technical University Munich in 2011. In 2016, Fabian received his PhD in computational biology from ETH Zurich, where he developed computational tools and probabilistic models for the analysis of genetic perturbation data.