Machine learning-powered Raman spectroscopy for rapid/timely/low-cost early diagnosis and monitoring of diabetic nephropathy

Jinfeng Ding, Zhengyi Fang, Xinjie Li, Lin Feng, Jiawen Shi, Xuxiang Zhou, Ping Huang, Hancheng Lin,
Machine learning-powered Raman spectroscopy for rapid/timely/low-cost early diagnosis and monitoring of diabetic nephropathy,
Microchemical Journal,
2025,
115738,
ISSN 0026-265X,
https://doi.org/10.1016/j.microc.2025.115738.
(https://www.sciencedirect.com/science/article/pii/S0026265X25030863)
Abstract: Diabetic nephropathy (DN) is a progressive complication of diabetes mellitus (DM), endangering the patient's life worldwide, yet technique in early identification of DN with high sensitivity and specificity is still absent. Given that the Raman spectroscopy-based approach allows for the rapid, low-cost analysis of molecular composition and variations, Raman spectroscopy was applied in assessing biochemical and metabolic alterations in the kidneys of DM mice at different stages. Preliminarily, Raman spectroscopy was proved to detect DM-stage-dependent (from DM-7w to DM-21w) biochemical and metabolic alteration in renal tissues, while the underlying changes in DM-7w mice could not be observed using hematoxylin and eosin (H&E) staining. Linear Discriminant Analysis (LDA) modeling of renal spectral data revealed distinct clustering patterns between early-stage DN and healthy mice, indicating the modality's potential for early diagnosis and pathological staging of DN. Critically, integration of Raman spectroscopy with optimized Convolutional Neural Networks (CNN)-guided machine learning classifier achieved high early DN diagnosis accuracy towards 97.16 %, contributing to a robust framework for both qualitative and quantitative assessment of DN. Future validation studies will focus on evaluating the technique's clinical-translational utility for early diagnosis and staging.
Keywords: Diabetic nephropathy; Raman spectroscopy; Machine learning; early diagnosis