Computational Imaging of Renal Structures for Diagnosing Diabetic Nephropathy
Summary – Proteinuria is the manifestation of a heterogeneous spectrum of kidney diseases that involve excessive loss of blood serum proteins to the urine. Damage to renal micro-compartments, including renal glomeruli, presents with proteinuria and may eventually lead to kidney failure. Medicare spends ~$24 billion annually on the care of >500K U.S. patients with end-stage renal disease. Kidney biopsies are often required to diagnose proteinuric renal disease. The traditional approach to diagnosing proteinuria includes qualitative or semi-quantitative manual estimation of glomerular structural damage in the renal biopsy. This process is approximate, subjected to user bias, time-consuming, and has low diagnostic precision in early disease stages. By projecting functional physiological measurements of glomeruli (e.g., eGFR, urine protein and serum creatinine levels) onto their histological structure, computational histological image analysis can more precisely identify structural changes that lead to physiological changes; this in turn reduces the required clinical resources and time for diagnosis, and provides clinicians with greater feedback on therapeutic efficacy. We have developed computational methods to analyze histological images of the heterogeneous renal microscopic architecture in proteinuric renal disease. In this proposal, we will analyze the performance of these methods to predict disease progression in human renal biopsies of proteinuric diabetic nephropathy (DN). We will computationally quantify morphologically diverse DN-indicative intra-glomerular features. We will analytically integrate computationally derived glomerular features with clinical biometrics in order to develop patient-specific disease quantification models. The innovation lies in the novel integration of traditional clinical detection methods with traditional diagnostic methods, under a computational schema that enables enhanced precision. This integration will lead to computational disease-predictive biomarkers of renal dysfunction in DN. We will investigate the predictive power of these markers to foretell future clinical endpoints from an earlier time point (i.e., DN progression from stage I to stage III or stage II to stage III). These methods support the development of quantifiable prognostic and predictive information, which is dynamic over the disease course, easily discriminated, and holds high informative power for modeling disease progression or response to therapy. This study will 1) enable earlier clinical predictions, thereby extending windows for interventions of evolving DN; and 2) work as a pilot platform for future studies to computationally derive renal biomarkers predictive of other diseases. Relevance – Automated detection of renal structural change with integration of physiological parameters will improve objectivity among clinicians, standardize renal diagnoses, and reduce time to a precise prognosis. Saving diagnostic time allows more patients to be treated in shorter amounts of time, reducing healthcare costs. Our tools will provide clinicians with invaluable quantitative information about their patient’s disease trajectory, enable identification of DN-specific signs earlier in disease progression, and extend windows of opportunity for early therapeutic interventions.

Going beyond segmentation in discovering early pixel level changes in diabetic nephropathy using histology image and urinary proteomics data
In the USA, approximately 34.2 million people have diabetes. Twenty-five percent of this population will develop diabetic nephropathy (DN). Diabetes is the most common cause of end stage renal disease (ESRD), accounting for approximately 45% of cases. In DN, manual scoring of renal microstructural damage often does not correlate with disease progression. Moreover, while glucose control is known to be the most acceptable treatment for diabetes, intensive control of glucose may not reduce the risk of DN. This gap in the literature prompts investigation of two overarching and interlinked questions in diabetes: (i) what are the early digital biomarkers that can be measured from histologically stained renal biopsies which predict DN and (ii) can we identify links between early image biomarkers and molecular pathways in DN. Answering these questions will help to better stratify DN patients into different risk categories, and aid precise strategic targeting of new and hopefully better treatment strategies for preventing or treating DN. The biggest investigative challenge is that there exist no open-source databases of well-curated digital images of DN renal biopsies with matching high throughput molecular data. While the generation of spatial transcriptomics or single cell/ nuclei RNA sequencing (scRNAseq / snRNAseq) data matched with renal biopsy of early DN cases is becoming technically possible, the sample sizes available from earlier efforts are still too low to answer the proposed questions. In this study, we will collaborate with nephropathologists, nephrologists, and molecular biologists to answer the proposed questions using matching renal tissue image and urinary proteomic (as molecular) data collected from patients with early and late DN stages. Our primary aim is to identify biological correlations between renal microstructural image patterns and urinary proteins in DN patients. Over the last six years, the PI has generated numerous results discussing objective quantification of renal micro-compartments, computational classification of renal diseases, and computational prediction of clinical biometrics from renal tissue images. In this study, the PI is refocusing his team’s efforts to investigate the molecular biomarkers from urinary proteomics and correlating/integrating the resulting information with matched histomorphometrics. This newly developed framework will answer the outlined questions pertaining to DN diagnosis and prognostication. The developed framework can be extended to conduct similar studies using other molecular omics data such as scRNAseq, snRNAseq, spatial transcriptomics, spatial metabolomics, and laser dissected spatial seq data.

Curation Flag Information