Abstract:Abstract Background and Aims Early diagnosis and evaluation play an important role in preventing the progression of diabetic kidney disease (DKD). Renal biopsy is the gold standard of DKD diagnosis. In 2010, the Renal Pathology Society (RPS) developed a consensus classification for DKD, which classifies DKD glomerular lesions and scores for tubulointerstitial lesion. However, as the pathologic heterogeneity of DKD patients and unparallel relationship between pathology features and clinical symptoms, it remains controversial whether is reliable to use RPS classification for renal outcomes prediction. Besides, the inconsistence between pathologists might magnify the gap between prediction and outcome. More reliable methods for DKD pathology assessment are needed. The development of machine learning (ML) algorithms especially convolutional neural networks (CNNs) enables it possible, which can provide automatic and precise analysis for complicated images. Thus, this study aimed to construct a ML-based model to assist and provide automatic decision supports to clinical doctors. Method We examined 204 DKD renal biopsied whole-slide images (WSIs) of Periodic Acid Schiff staining from Southeast University Institute of Nephrology whose database came from more than 30 hospital centers. All the WSIs were randomly annotated by three pathologists. We established a consecutive CNNs framework: DKD pathology evaluation system (DPES). It consists of three parts: 1. Based on classic UNet architecture, we developed a segmentation model whose backbone is replaced by ResNet 50 architecture for detecting glomeruli. 2. According to RPS classification, a CNN added with channel attention mechanism is built to discriminate sclerotic glomeruli, glomeruli with K-W node, and glomeruli with mesangial expansion. Through this method, the entire WSIs were divided into RPS grades I-IV. 3. For the analysis of tubulointerstitial lesion, we constructed a segmentation model as previous description to calculate the area of the atrophic tubule and its percentage of the total tubules. Results For glomerular assessment, compared with other ML networks, DPES achieved better performance in segmentation for glomeruli (Intersection over Union (IOU): 0.82, Precision: 0.89, Accuracy: 0.91) and tubulointerstitial (atrophy (TA): IOU: 0.9;Precision: 0.94; Accuracy: 0.95; non-tubular atrophy (NT): IOU: 0.84; Precision: 0.90; Accuracy: 0.93). Interstitial fibrosis and tubular atrophy (IFTA) from network compared with three pathologists didn’t show statistically difference. Conclusion Our findings demonstrated that the DPES achieved similar performance with three pathologists. It suggests that machine learning algorithms can generate reliable results and provide more supports for clinical decision.

Spatial-Spectral Density Peaks Based Discriminant Projection for Classification of Membranous Nephropathy Hyperspectral Pathological Image1

Hyperspectral pathology image classification using dimension-driven multi-path attention residual network

A Spectral-Texture Kernel-Based Classification Method for Hyperspectral Images.

Semi-supervised Enhanced Discriminative Local Constraint Preserving Projection for Dimensionality Reduction of Medical Hyperspectral Images

Classification and quantification of glomerular spike-like projections via deep residual multiple instance learning with multi-scale annotation

Accurate classification of glomerular diseases by hyperspectral imaging and transformer

Multiplicity of solutions for a class of quasilinear elliptic equation involving the critical Sobolev and Hardy exponents

Cell Classification Using Convolutional Neural Networks in Medical Hyperspectral Imagery

Rapid diagnosis of membranous nephropathy based on serum and urine Raman spectroscopy combined with deep learning methods

Mask-Guided Spatial–Spectral MLP Network for High-Resolution Hyperspectral Image Reconstruction

Deformable R-CNN for Detection of Electron Dense Deposits in Glomerular Transmission Electron Microscopy Images

Spectral Machine Learning for Pancreatic Mass Imaging Classification

A 3-stage Spectral-spatial Method for Hyperspectral Image Classification

#2775 USING A DEEP LEARNING MODEL TO EVALUATE PATHOLOGICAL INJURY OF DIABETIC KIDNEY DISEASE

Heterogeneous Spectral-Spatial Network with 3D Attention and MLP for Hyperspectral Image Classification Using Limited Training Samples

Pseudo Complex-Valued Deformable ConvLSTM Neural Network With Mutual Attention Learning for Hyperspectral Image Classification

Blood Cancer Diagnosis Using Hyperspectral Imaging Assisted with Forward Searching Method and Machine Learning

Joint Diagnostic Method of Tumor Tissue Based on Hyperspectral Spectral-Spatial Transfer Features

Diagnosis of Cholangiocarcinoma from Microscopic Hyperspectral Pathological Dataset by Deep Convolution Neural Networks.

Hybrid Convolutional Network Combining Multiscale 3D Depthwise Separable Convolution and CBAM Residual Dilated Convolution for Hyperspectral Image Classification

Hyperspectral Image Classification Based on 3-D Octave Convolution With Spatial–Spectral Attention Network