P263 Raman spectroscopy analysis of saliva combined with an artificial neural network algorithm could discriminate between Ulcerative Colitis and Crohn’s disease

Abstract

Abstract Background Early and accurate detection of inflammatory bowel disease (IBD) is crucial for proper intervention.The IBD diagnostic process can be both time-consuming and invasive and lacks a gold standard in disease differentiation. Saliva analysis has gained interest as a potential non-invasive source of disease biomarkers due to its positive correlation with blood-based constituents (e.g. interleukins) at a molecular level. We systematically studied the molecular fingerprint of saliva and tissue biopsy, including how they compare as well as changes associated with disease state-IBD vs non-IBD and Ulcerative colitis (UC) vs Crohn’s disease (CD). Methods After obtaining informed consent Saliva and colon biopsies were collected from patients undergoing endoscopy as a standard of care procedure at University Hospitals Birmingham, NHS Foundation Trust, Birmingham, UK. The samples were correlated with their tissue pathology markers and findings of the endoscopic evaluation. By combining Raman spectroscopy and an artificial neural network algorithm, Self-Optimising Kohonen Index Network (SKiNET), we have developed a non-destructive molecular profiling tool that assesses both salivary and colonic biopsy changes, which helps to determine the diseased state of the sample source. The collected Raman spectra were analysed using SKiNET as a decision support tool to identify spectral markers and discriminate between healthy and IBD disease classes [1]. Results This pilot study characterised the spectral signatures of 101 saliva samples (healthy = 50, UC = 27 and CD =24) and 44 tissue biopsy samples (healthy = 12, UC = 17 and CD = 15). The algorithm reached 87.5% sensitivity and 89% specificity when classifying saliva and 83.9% sensitivity and 84.6% specificity when classifying tissue biopsy. When comparing the spectral fingerprints of saliva to tissue biopsy, the UC spectra are of similar forms with comparable intensities at each of the primary peaks; 1003 cm1(phenylalanine), 1440 cm-1 (Lipids) and 1656 cm-1 (Amide I). In contrast, CD saliva and tissue biopsy indicate significant variance throughout the Raman spectral fingerprint region, with intensity changes observed at each of the primary peaks and the presence of additional shoulder peaks associated with proteins and lipids. In identifying markers of disease IL-4, IL-8 and IL-10 were identified as having important and designatory Raman bands which could serve in the discrimination of IBD in individuals. Conclusion This unique combination of Raman spectroscopy and advanced machine learning represents significant progress towards improved, non-invasive and rapid classification of IBD.This technique also lays the foundation for future applications in disease diagnosis, monitoring and therapeutics.