Non-invasive diagnosis of Crohn's disease based on SERS combined with PCA-SVM

Abstract

Crohn's disease (CD) is an idiopathic chronic inflammatory bowel disease without a cure. Most of the CD patients are firstly diagnosed by invasive endoscopy, and clinical and pathological examinations are further required to confirm the diagnosis. Hence, the development of a non-invasive, rapid and accurate diagnosis method for CD patients is essential. In this study, urine samples from 95 CD patients (including 58 active CD (aCD) patients and 37 inactive CD (iCD) patients) and 48 healthy controls (HC) were investigated by surface-enhanced Raman spectroscopy (SERS). The statistical analysis of the three groups (i.e., CD/HC, aCD/HC and iCD/HC) was performed on the measured data. Principal component analysis (PCA)-support vector machine (SVM) and PCA-linear discriminant analysis (LDA) were then employed to establish classification models to distinguish between patients and HC. For the average SERS spectra of patients and HC, the Raman peaks belonging to lipids, proteins and nucleic acids were stronger in patients than those in HC. It showed that the classification accuracy of CD/HC based on PCA-SVM was higher than that of PCA-LDA (82.5% vs. 69.9%). And the classification accuracy of aCD/HC based on PCA-SVM was higher than that of iCD/HC (86.8% vs. 76.5%). The classification model we established distinguished between aCD and HC with 86.2% sensitivity and 87.5% specificity. It indicates that the metabolic change of patients could be identified by measuring urine with SERS, and aCD and HC could be distinguished more effectively. Our findings are helpful for clinicians to diagnose CD patients and monitor the progress and recurrence of the disease.