Testing the hypothesis of the universality of the biochemical basis of the skin in the diagnosis of chronic kidney disease using Raman spectroscopy and artificial intelligence

Abstract

In this study, the hypothesis of the existence of a universal biochemical basis of the skin was tested: whether changes in the relative content of the same basic components (keratins, collagen, lipids, melanin, water, and their mixtures) are informative for both the diagnosis of dermatological diseases and the detection of systemic pathology, such as CKD. MCR-analysis was used to isolate biochemical components using an eight-component set previously isolated from the skin spectra of patients with dermatological diseases. The classification was performed using three algorithms: random forest, gradient boosting, and multilayer perceptron with 10-fold cross-validation. As a result of the study, the hypothesis was confirmed, and the borrowed dermatological basis was successfully applied for the diagnosis of CKD. The feature importance analysis showed that all models consistently identified a mixture of proteins, lipids, natural moisturizing factor, and melanin as the most informative marker of systemic metabolic imbalance. The best results were demonstrated by a random forest after limiting the depth of the trees and excluding non-specific optical contributions: the ROC AUC reached 0.94±0.04 with a sensitivity of 95.6% and a specificity of 86%. Thus, the proposed technique allows for a highly accurate non-invasive diagnosis of the terminal stage of CKD, confirms the existence of universal biochemical markers in the skin that change in response to a wide range of pathological conditions, and substantiates the potential for creating a unified diagnostic tool based on the analysis of the molecular composition of the skin.

Speaker

Tomnikova Ksenia
Samara National Research University
Russia

Discussion

Ask question