Brain Digital Slide Archive: An Open Source Platform for data sharing and analysis of digital neuropathology

点击翻译按钮获取中文摘要

Recent advances in machine learning and computer vision have had transformative effects on the medical imaging field. Algorithms can now automatically identify patterns and objects in images, often with a degree of precision rivaling human experts for certain tasks. Key to these advances is the availability of large, well-curated datasets in machine-readable formats. Neuropathologic evaluation of brain tissue is central to the diagnosis and staging of Alzheimer's Disease (AD) AD and Related Dementias (AD/ADRDs) but the underlying histology data is not widely and easily shared. The increasing availability of whole slide imaging systems now makes the distribution of histologic data simpler and enables image analysis algorithms to be developed and applied, but numerous barriers exist before such technology can be widely adopted by the neurodegenerative research community. The lack of standard file formats and naming schemas, ensuring subject privacy, subject de-identification, and the enormous size of these images are ongoing challenges. Through NCI/NIH U24 and U01 grants focused on cancer-related image analysis workflows, we have previously developed the Digital Slide Archive (DSA). In this project, we propose to enhance the DSA platform with functionality geared specifically for the neurodegenerative neuropathology community, creating a federated open-source Brain Digital Slide Archive (BDSA) platform. The BDSA is designed to allow the seamless sharing of imaging data, annotations, and metadata amongst participating sites, and to enable the training and deployment of image analysis algorithms on multi-institutional data sets. This includes developing a standardized data dictionary to describe slide-level metadata, and tooling to facilitate data cleanup. We will test these tools and infrastructure by conducting various proof of principal analysis workflows. These include the ability to centrally discover and annotate images stored in geographically distinct regions and run algorithms to identify neurofibrillary tangles (NFTs) using slides from 4 distinct geographic sites (Emory University, University of California Davis, University of Pittsburgh, and Northwestern University) digitized using multiple scanner models and manufacturers. The system will also allow users to securely transfer images to a central location, which may be necessary for certain analytic workflows. These objectives, paired with our complimentary and synergistic expertise in informatics, neuropathology, and engineering, will aid in the development of robust, scalable, reliable, and shareable platforms to provide a foundation for innovative and transformative science addressing a critical unmet need in AD/ADRD research.