Cloud based digital pathology repositories for sharing WSI to build and validate AI models
Biorepositories and institutional databanks store a large amount of diverse specimen data with the goal of allowing researchers to select relevant and representative specimens, minimize tissue loss, and promote the sharing of knowledge between institutions to produce meaningful translational biomedical research. The advent of digital pathology has presented a unique opportunity to derive additional value from these biorepositories by allowing sharing of imaging data for research activities. Given the quantum of research being currently conducted with digital slides within institutions, AI companies, pharma and industry, these databanks can play an extremely important role for distribution. However, there are no well established, standardized workflows that allow easy distribution of data, ability to collaborate with other researchers, or intuitive tools that can be used by researchers when working with whole slide images (WSI) for research.
Recently, the Cooperative Human Tissue Network (CHTN) Midwestern division and PathPresenter have established a collaboration to explore the creation of a next generation tissue repository. The objective of this project was to create an elastic compute cloud based infrastructure, scalable workflows, and make the data easily available to researchers as well as provide “smart” tools to maximize collaborative research, to ultimately use the data to benefit patients and donors.
To enable an innovative digital biorepository, a web portal was created that was owned and managed by CHTN administrators. Over the span of 6 months over 8,000 WSI from diverse organ systems were uploaded and associated with de-identified tissue metadata including diagnosis, tumor grade, relevant treatment, outcome data, etc. Administrative tools were developed to allow researchers from anywhere around the world to register for access. After appropriate vetting and clearances from CHTN administrators, users were granted access to search and leverage data contained within the data repository. To produce meaningful research, deep search capabilities allowed researchers to enter detailed search criteria to identify specific datasets that were relevant to their research. Data could be downloaded or aggregated in private folders to use for online research including user defined regions of interest, annotations, metadata, and even the deployment of AI modules. With functionalities to delineate users roles and permissions (e.g. view/edit/delete annotations) and tools to audit annotator progress; researchers using these web-based tools can build entire research projects on the platform, as well as test AI models without the need to exhaust a single biospecimen.
In the presentation we will present a model for creating an infrastructure for biorepositories and institutions to share data in a secure and compliant ecosystem as well as provide analytical tools to maximize the use of the data. Tools to guide biospecimen selection for genomic studies, as researchers can easily visualize WSI on the platform before selecting the block with maximum tissue and tumor. Incorporating the ability to upload other omics data (Radiology DICOM files, genomic data and tissue arrays) as well as clinical and outcome data, allowing the derived knowledge to feedback into the biorepository and truly enhance research capabilities that will benefit patient care will be presented.
- Discuss tools for biorepositories to allow selection and sharing of relevant and representative WSI with multiple researchers
- Discuss “SMART” tools for creating annotations and integrating AI models for use in research
- Connecting WSI images with other omics data, clinical and outcome data can truly enhance future research
Rajendra Singh, MD
Director of Digital Pathology and Dermatopathology
Anil Parwani, MD, PhD, MBA
Professor of Pathology
The Ohio State University