Digital Pathology and Computational Image Analysis for Lupus Nephritis

Abstract Nearly one third of all adults with Systemic Lupus Erythematosus (SLE) present with lupus nephritis (LN) at diagnosis and up to two thirds manifest kidney involvement during the course of the disease with higher prevalence in minority populations particularly those of African ancestry. Furthermore, as many as 10% of patients with LN will progress to end-stage and require dialysis and/or kidney transplant. The ability to accurately identify LN patients at risk for progression could shift much of the current management paradigm from treatment to prevention. However, the prognostic significance of histopathologic classification of LN, the most current arising from a collaboration between the International Society of Nephrology and the Renal Pathology Society (ISN/RPS) in 2004 is controversial. Therefore, novel approaches are required to obtain continuous, quantitative data to improve accuracy, reproducibility, and prognostic utility. During the last decade, the rapid evolution of digital image technology has begun to challenge the established light microscopy-based protocols. Digital pathology, a dynamic, image-based environment for the acquisition, management, and analysis of information generated from digitized images, is emerging in the setting of clinical trials and research, including kidney disease, but has yet to be comprehensively applied to lupus nephritis biopsy interpretation in a large cohort. We hypothesize that digital pathology and image analysis approaches will improve the prognostic utility of the kidney biopsy in lupus nephritis and allow more efficacious treatment approaches. Thus, the goal of our proposal is to take advantage of the recent advancements of digital pathology, deep learning, and machine learning, both supervised and unsupervised methods, and the broad and extensive experience of the research team, and apply 1) quantitative morphology, 2) supervised computer- aided deep learning image analysis algorithms with morphometric analysis, and 3) unsupervised machine learning algorithms. To test our hypotheses, we will use our Digital Pathology Image Repositories (DPIR). We have generated our DPIR from approximately 550 whole-slide scans of lupus nephritis biopsies from X patients enrolled in the Accelerated Medicines Partnership (AMP) for Lupus Nephritis and the University of Michigan Lupus Cohort. Data from the AMP cohort includes extensive clinical parameters obtained during longitudinal follow up for 52+ weeks with predefined clinical outcomes of renal response. The Michigan Lupus Cohort likewise includes detailed clinical information over several years already.