SSR – A Searchable, Sharable, and Reproducible Imaging Informatics Environment for Biomedical Image Storage, Annotation, and Analysis
Event Type
TimeTuesday, July 306:30pm - 8:30pm
LocationCrystal Foyer and Crystal B
DescriptionApplication of machine learning (ML) and artificial intelligence to biomedical images has huge potential to improve treatment of cancer and other diseases. Classification and segmentation of histopathology and radiology images is critical to diagnosing cancer in clinical settings and understanding cancer in research. Yet one great barrier to applying ML to this problem is the quality and quantity of annotated data for training models. Deriving informatics from images requires quantification and thus segmentation, a time consuming and laborious process for subject-matter experts. We have created the Searchable, Sharable, and Reproducible (SSR) Imaging Informatics environment, a research computing platform, for managing, annotating, and analyzing biomedical images through an intuitive and accessible web interface to overcome this obstacle. Integrated annotation and analysis workflows in SSR allow for updates to manual annotations with model annotation predictions in situ. SSR is based on the established, open-source data platform Girder (Kitware) with data models stored in MongoDB and files on local file systems, HDFS, and Amazon S3. In-browser, JavaScript-based gigapixel image and volume viewing and pixel-level annotation tools synchronize with servers through RESTful APIs. Distributed analysis of the annotated data is managed by Celery task queues capable of launching Docker container instances for ease of portability on high-performance computing clusters and utilization of current imaging processing and ML libraries with GPU optimizations. SSR provides easy to use, web-based pixel annotation tools combined with integrated ML image segmentation to greatly enhance production and management of training data for ML models.

Funded by NCI Contract No. HHSN261200800001E