Deep learning approach for the assessment of arterial and arteriolar lesions in kidney biopsies and prediction of cardiovascular risk

PROJECT SUMMARY / ABSTRACT Chronic kidney disease (CKD) is a major public health burden and affects an estimated 10% of the population worldwide. CKD is also a major risk factor for cardiovascular disease (CVD), hospitalization, and premature death. There are only few therapies available to reduce remission, but progression of the disease and its

vascular complications can be delayed with access to timely and accurate diagnostics and early treatment. Without diagnostic pathology assessment of kidney biopsy tissue, the underlying disease process and extent of microvascular damage are frequently unknown in an individual patient as the main measures used to

diagnose and stage CKD – the estimated glomerular filtration rate and proteinuria – do not provide insight on underlying histopathologic changes and have limitations related to specificity and diagnostic accuracy. However, pathologist assessment of biopsy tissue is costly, impacted by inter-rater variability, and not available

for many patients at a global level. Computer-vision, a field of artificial intelligence (AI) that uses computer systems to derive meaningful information from digital images, can help to overcome these limitations. In this proposal, we will utilize an advanced form of machine learning (a branch of AI that allows computers to learn without being explicitly

programmed) known as deep learning (artificial neural networks that search for common characteristics to organize data) to associate image-derived features from digitized kidney biopsies with clinical phenotypes. We will perform the following tasks: 1) develop and validate an automated assessment of kidney vascular

pathology and risk of adverse cardiovascular events, and 2) integrate the information from digital pathology, clinical parameters, and CKD proteomics (7,000+ proteins) into a larger prediction framework. Deep learning algorithms will be developed using the Boston Kidney Cohort, a prospective cohort study of individuals with

biopsy-confirmed kidney disease, and validated in the Kidney Precision Medicine Project and the Boston Medical Center CKD Cohort. The Principal Investigator of this study, Dr. Insa Schmidt, is an early-stage career investigator with prior research experience in kidney disease epidemiology, biomarkers, and -omics analysis.

With this K01, Dr. Schmidt plans to expand her research beyond biomarkers and epidemiology to encompass applied data science, to be able to analyze far larger datasets that are expected to play a major role in clinical research and nephrology. Dr. Schmidt will be supported by an interdisciplinary mentorship and advisory team,

comprising experts in data science, kidney pathology, omics integration, and biostatistics. Training in these fields will be facilitated through a combination of formal coursework, seminars, and mentored research experience. Completion of the research and training proposed in this K01 will position Dr. Schmidt to reach her

long-term career goal to become an independent investigator with expertise in combining epidemiologic tools with machine learning techniques to improve the lives of patients with kidney and cardiovascular disease.