Classification of Glomeruli with Membranous Nephropathy on Renal Digital Pathological Images with Deep Learning
Hao Fang, Ming Li, Xueyu Liu, Xinyu Li, Junhong Yue, Weixia Han · 2020
Membranous nephropathy (MN) is the one of the most common pathological types that cause adult nephrotic syndrome (NS). Recently, the incidence of MN has shown a clear upward trend. Nevertheless, there is no more accurate and fast artificial intelligence algorithm for diagnose of MN which work is laborintensive and time-consuming if it is done manually. In this article, MN-Net, a CNN-based method, is applied to glomeruli detection and classification on whole slide images (WSIs). This work is mainly divided into two parts, a glomerulus detection network and a classification network. The detection network is utilized to locate glomeruli on WSIs. Multiple instance learning (MIL), a weakly supervised classification network following detection network classifies the glomeruli detected earlier. Our network is training on PASM-stained WSIs of 1281 cases collected from multi-centers. Experimental results prove that our method is effective with a high precision of 99.66% for glomeruli detection and 99.53% for MN glomeruli classification on this dataset. In summary, this method has been proved to be an effective method with advantages of speed, high accuracy, strong robustness, and low cost of data annotation that can be applied to the diagnosis of renal pathology. In the future, this method can also be extended to the classification of other glomerular diseases under light microscope (LM). The introduction of glomerular basement membrane (GBM) segmentation and measurement models can further improve the reliability of this model.