Artificial Intelligence-based & Ethically-focused Multi-Modal Data Integration Framework for Screening, Diagnosing, and Caring for CKD Patients

Abstract We will develop a distributed, cloud-based, artificial intelligence (AI) tool for the prediction of chronic kidney disease (CKD) progression and modifiable risk factors for clinical use. Our tool will provide a 5-year risk profile individualized to each patient and will integrate electronic health record (EHR) data, text data, radiology images with corresponding reports, and pathology data, including brightfield histology images and electron microscopy reports. The EHR, text, laboratory, and radiology data from >200K individuals with kidney disease will drive the AI tool development supplemented with histopathologic image data from >1000 individuals. Although histopathologic information has long held prognostic significance for end-stage renal disease, such data is rarely included in clinical kidney failure risk equations due to its complexity. Because the histopathologic manifestations of CKD are broadly distributed across the kidney’s functional units, our tool will facilitate the decomposition of each kidney whole slide image into hundreds of sub-images, comprising >~3M functional units, quantitate pixel-level features, and provide balance among the different modalities for effective learning and generalization. The resulting tool will support: primary care physicians and nephrologists who will benefit from enhanced risk prediction using complex data types, patients who will benefit by identification and prioritization of modifiable risk factors, and nephropathologists who will benefit from the quantitative analytics. An iterative process will incorporate their feedback to enhance the usability and usefulness of our platform. Our platform will be self-adapting, and will ingest and learn from new data as it evolves, improving kidney disease trajectory prediction and therapeutic care.