Intensity grading of kidney biopsy direct immunofluorescence IgG image via semantically enhanced feature network

Abstract

The intensity of direct immunofluorescence IgG imaging for kidney biopsy is an important reference index for the diagnosis of common chronic kidney diseases (CKD) such as membranous nephropathy (MN). Although this index is widely used in diagnosing membranous nephropathy, the intensity levels are currently judged only by the subjective perception of fluorescence luminance in pathological images through physicians’ naked eyes. Therefore, consistent intensity level classification cannot be achieved. To address this issue, this paper proposes a Preprocessing Semantically Enhanced Feature image classification method. The segmentation network is used to segment the original kidney biopsy direct immunofluorescence IgG images to isolate the diseased glomeruli. These segmented glomeruli images are then compared with the original glomeruli images to standardize the kidney biopsy direct immunofluorescence images. Simultaneously, considering the characteristics of small inter-class differences and large intra-class differences in fine-grained images, a semantically enhanced feature image classification model is employed to automatically classify the intensity of the standardized images. The method was evaluated on a constructed real clinical dataset, and the experimental results demonstrated that the proposed method achieved high classification accuracy on this dataset. The source code is made available on https://github.com/SnowRain510/PSEF-NET