Deep Learning Enables Automated Segmentation and Quantification of Ultrastructure from Transmission Electron Microscopy Images

The widths of kidney glomerular basement membrane (GBM) and podocyte foot processes (FP) are essential ultrastructural markers for assessing kidney function and diagnosing glomerular disease. FP widening reflects podocyte injury, whereas GBM thickening is characteristic of conditions such as diabetic nephropathy and Alport syndrome. Current measurement practices depend on manual tracing and expert review, which are labor intensive, variable between observers, and difficult to reproduce. We present an automated deep learning dbased framework TEAMKidney for accurate and scalable measurement of GBM and FP widths in transmission electron microscopy (TEM) images of glomeruli. The framework combines a proposed Glom2Mask model for panoramic segmentation, enabling robust identification of GBM and FP across species, including mouse, rat, and human kidney samples. Segmentation outputs are processed by computational pipelines to extract and quantify GBM and FP widths. The proposed method showed high concordance with expert annotations, substantially reduced measurement time, and improved reproducibility across datasets. This framework supports both experimental research and clinical applications, offering a reproducible and efficient approach to ultrastructural kidney analysis. By reducing dependence on manual measurements, it provides a scalable digital pathology solution with direct relevance to nephrology research, diagnostic practice, and clinical trials.