Machine learning-assisted SERS-based profiling of blood serum filtrate fractions in cervical cancer using 50 kDa filtration devices

Persistent infection with human papillomavirus (HPV) (HPV-16 and HPV-18) is the main cause of cervical cancer, and it continues to be one of the common causes of cancer-related deaths in women globally. To detect this cancer at an early stage, several diagnostic methods have been explored, but still a need for an easy-to-perform, non-destructive, and non-invasive method that can detect this cancer with high sensitivity in a short time. In this study, SERS was used to detect and classify cervical cancer using blood serum filtrate fractions collected with 50 kDa ultrafiltration devices. In filtrate fractions, prominent SERS spectral peaks were recorded at 326, 345, 883, 927, 1102, 1133, 1203, and 1403 cm-1 which could be used as potential cervical cancer biomarkers. As a SERS substrate, silver nanoparticles (Ag-NPs) were used to enhance spectral resolution and the intensity of weak Raman signals. SERS spectra of serum samples from cervical cancer and healthy patients were analyzed by principal component analysis. The four-class PLS-DA model was subsequently developed using sample-level partitioning to distinguish healthy controls from stage 2, stage 3, and stage 4 cervical cancer samples. Sample-level LOOCV identified eight latent variables as the optimal model complexity, with a maximum LOOCV accuracy of 81.82%. Independent sample-level validation yielded an overall classification accuracy of 86.67%, while one-versus-rest AUC values were 1.000, 0.909, 0.955, and 1.000 for healthy, stage 2, stage 3, and stage 4, respectively.