Automated quantification of ultrastructural pathology of diabetic nephropathy using deep learning
Behzad Najafian   (Seattle, WA)
Diabetic nephropathy is by far the most common cause of end stage kidney disease. The combination of morphometric approaches and electron microscopy have provided major contributions to the current understanding of the progression of diabetic nephropathy. Structural changes when properly quantified are not only regarded as robust biomarker of progression and severity of diabetic nephropathy, but also correlate with renal function and can predict progression of diabetic nephropathy. Application of these methods has been limited to research studies, largely because currently automated approaches are not available. Deep learning is a form of machine learning methods based on artificial neural networks which has been proven to be a powerful tool for image analysis. Here, we aim to develop deep learning algorithms to automate segmentation and quantification of key glomerular structural parameters that are relevant to diabetic nephropathy, including glomerular basement membrane thickness, podocyte foot process width and expansion of mesangium and mesangial matrix. We will train deep learning algorithms on a large collection of electron microscopy images from kidney biopsies obtained from patients with type 1 and type 2 diabetes with a wide spectrum of diabetic nephropathy severity, as well as kidney biopsies from normal control subjects and will validate the method using multiple approaches.
Data for this report has not yet been released.