Non-Invasive Precise Classification of Glomerular Diseases in Urine Based on Hyperspectral Technology

Area of Science:

• Nephrology
• Medical Imaging
• Artificial Intelligence

Background:

• Glomerular diseases represent a significant global health challenge.
• Current diagnosis relies on invasive renal biopsy.
• There is a need for non-invasive diagnostic tools.

Purpose of the Study:

• To evaluate hyperspectral imaging (HSI) as a novel non-invasive diagnostic method for glomerular diseases.
• To develop and validate a machine learning model for classifying glomerular disease subtypes using HSI data.

Main Methods:

• Urine samples from patients with four glomerular disease subtypes (Minimal Change Disease, Diabetic Nephropathy, Membranous Nephropathy, IgA Nephropathy) were analyzed using HSI.
• A ResNet-50 classification model was developed using dimensionality-reduced HSI spectral data.
• The model underwent five-fold cross-validation for performance assessment.

Main Results:

• The ResNet-50 model achieved 96.8% average cross-validation accuracy.
• The model demonstrated a high area under the curve (AUC) of 0.982.
• Accurate multiclass differentiation of glomerular disease subtypes was feasible with limited samples.

Conclusions:

• Hyperspectral imaging combined with a ResNet-50 model is a highly effective non-invasive approach for diagnosing glomerular diseases.
• This integrated methodology shows superior efficacy compared to traditional methods for classification.
• The study confirms the feasibility of using HSI for early and accurate detection of kidney disease subtypes.