Glomerular Classification of Membranous Nephropathy Based on Deep Residual Learning with Large Resolution Images

Membranous nephropathy (MN) is the leading cause of nondiabetic idiopathic nephrotic syndrome in adults world-wide. The pathological changes of the glomerulus are taken as the “gold standard” for the diagnosis of MN.We use the classical deep residual network ResNet as the base network to classify the glomeruli. Because the size of the obtained glomerular image can reach more than 600*600 pixels, the size of the image input to the network is increased to 672*672 to reduce the loss of image information. We add two residual downsampling blocks to reduce the size of the feature maps and further extract features. At the same time, the attention mechanism can pay attention to the features that are important for classification of high similarity images, so we also introduce the attention mechanism. Since normal glomeruli are challenging to obtain, we used minimal change disease(MCD) glomeruli to substitute normal glomeruli. For classification of the MN glomeruli and MCD glomeruli, our model achieved an accuracy of about 87.5%. Our proposed deep residual network has demonstrated promising results in processing glomerular images with higher pixel resolution in the metaverse. With further refinement and validation, this model holds potential for applications in clinical research, providing valuable assistance to physicians.