Federated Digital Pathology Platform for AD/ADRD Research and Diagnostics

Abstract We will connect multiple Alzheimer’s disease and related dementias (AD/ADRD) research centers for optimized and standardized whole slide image (WSI) advanced analytics. We propose these two Specific Aims: Specific Aim 1: Generate a federated platform for data sharing and analysis of human digital neuropathological (DNP) slides. Specific Aim 1a: Develop an open-source platform to aggregate metadata distributed across multiple repositories into a central registry portal. This subaim will follow federated data curation, harmonization, annotation, and standardization employing FAIR (findable, accessible, interoperable, and reusable) principles. Specific Aim 1b: Develop a federated data curation and management system. This subaim provides methods to curate physical data, such as slides for scanning, and existing digital data across federated sites. WSIs will be shared with private repositories and high-performance clusters. Specific Aim 2: Develop and demonstrate a platform for federated machine learning/artificial intelligence (ML/AI), result evaluation, and central project information and management hub for AD/ADRD studies. Specific Aim 2a: Develop a federated data annotation and automated AI/ML processing system. This subaim provides methods for users to prepare AI-ready multimodal (image, genomic, text, etc.) datasets from distributed sources and conduct multi-site data transfer and federated training. Services and tools developed in Aim 1 will be used to programmatically acquire, generate, and populate provided and user-defined AI/ML pipelines. Specific Aim 2b: Develop an open-source platform for the review, evaluation, and promotion of datasets and associated models. This subaim will provide a project evaluation portal integrating cohort, dataset, and model training results within a unified interface. A public-facing model hub will be developed to provide project, data, specifications, evaluation, and model data to support reproducibility efforts, data sharing requirements, and promote projects. Five separate projects are proposed. Common themes include the study of tau/TDP-43opathies and a diverse sampling of human populations to leverage the unique strengths of DNP. Each of the following sites will contribute resources, expertise, experimental study design, data, and analyses: University of Kentucky (PIs Nelson and Bumgardner, Co-Is Cheung and Fardo). Generate standardized SOPs useful for DNP diagnoses and research along with ML/AI expertise for a federated network that leverages expertise from cancer field. Northwestern and UTSA Universities (PI Flanagan). A focus on TDP-43 proteinopathy in community-based cohorts and ethnoracially diverse populations. University of São Paulo and UCSF (Co-Is Suemoto and Grinberg). Evaluate ADNC from brains of individuals with African and non-African ancestries. University of Toronto (Co-I Kovacs): Using ML to segment gray vs white matter pathologic evaluations of tauopathies. University of Washington (Co-I Keene): Tau pathologies across age and environmental exposures with a focus on traumatic brain injuries.