SERS spectrum of saliva combined with machine learning algorithm for early diagnosis and monitoring of periodontal disease.

To achieve non-invasive early diagnosis and severity monitoring of periodontal disease, this study employed silver nanoparticles as a surface-enhanced Raman Scattering (SERS) substrate for the detection and analysis of salivary SERS spectra from a control group (periodontal health group) and groups with periodontal disease of varying severities (including gingivitis, as well as mild, moderate, and severe periodontitis). The results demonstrated significant differences in salivary SERS spectra between the control group and the gingivitis group, between the control group and the mild periodontitis group, and between the control group and groups with periodontal disease of different severities. Subsequently, based on the dual screening criteria of Variable Importance in Projection (VIP) ≥ 1 and P < 0.05 in the partial least squares discriminant analysis (PLS-DA) model, potential biomarkers related to disease status, such as glycogen, ascorbic acid, uric acid, hypoxanthine, glutathione, D-mannose, and phenylalanine, were identified. Finally, a convolutional neural network (CNN) model was applied for classification, achieving an average accuracy of 99.70 % in distinguishing the control group, gingivitis group, and mild periodontitis group, and 99.43 % in differentiating the control group from periodontitis groups of different severities. The implementation of this method is expected to be applicable to the non-invasive early diagnosis of periodontal disease and the monitoring of its severity.