Deep Learning Segmentation of Kidney Tissue Microarrays Using Infrared Spectral Imaging

Abstract

Renal function is an essential marker in the classification of renal disease and clinical symptoms of renal failure develop when there is 15% renal function. In this study, we used infrared spectroscopic (IR) imaging to investigate biomolecular markers from renal transplant biopsies. These images are used for the classification of regions of fibrosis from biopsies containing renal cell carcinoma (chromophobe and oncocytoma) and the prediction of fibrotic proliferation using biochemical signatures. IR spectroscopy is a diagnostic approach utilizing human tissue to label biochemical signatures. Images are captured in several hundred wavelengths in the infrared region of the electromagnetic spectrum giving researchers access to more information than traditional RGB images captured by a microscope. While images captured in several bands are great for disease diagnosis, it poses significant challenges for manual cell review by a pathologist. Our project goals are to apply feature selection to remove data with less importance and reduce dimensionality. We also hope to apply a deep learning model on filtered dataset for identification of fibrosis.