Team 7: Deep learning cell annotation framework for tumor classification

Deep learning is poised to dramatically transform the process of characterizing diseased tissue, a primary driver of clinical decisions. However, the massive size of individual tissue images poses a major obstacle, both in the generation of training data and the evaluation of classifier output. While specialized schemes exist for displaying these large gigapixel images, we lack similarly efficient approaches to visualize and edit the massive associated annotations produced by the classifiers.

Here we propose to build an efficient annotation storage/visualization/editing framework that can be coupled with existing web-frameworks (e.g. openslide+openseadragon) to visualize tissue images. You will be provided gigapixel tissue images and trained classifiers that segment individual nuclei in them. Your goal is to build a web-framework that will enable 1) smooth zooming/panning of >100,000 annotations overlaid on these gigapixel images and 2) editing of selected annotations. Participants are encouraged to build on design choices made by existing frameworks (e.g., with similar goals albeit for much smaller number of annotations.

Team Lead: Satwik Rajaram, Bioinformatics & Pathology,