Machine learning-assisted SERS for rapid and accurate early screening of cervical cancer

Early and accurate screening of cervical cancer (CC) is critical for improving patient prognosis and reducing mortality, and there is an urgent clinical demand for noninvasive, sensitive, and reliable detection methods. In this study, a serum detection platform based on surface-enhanced Raman scattering (SERS) was constructed, and a principal component analysis-multiple kernel support vector machine-grid search (PCA-MKSVM-GS) model was established to realize accurate and stable identification of cervical cancer patients. High-performance Ag nanocube arrays (AgNC arrays) were fabricated as SERS-active substrates to effectively acquire serum molecular fingerprint information. Serum SERS spectra were collected from cervical cancer patients and healthy subjects, and subsequently subjected to spectral preprocessing including noise reduction, baseline correction, and normalization. The preprocessed spectra were imported into the model, where PCA performed feature extraction and dimensionality reduction, MKSVM enhanced the nonlinear fitting ability for complex spectral features, and GS automatically optimized the hyperparameters of the model. The accuracy evaluation, feature visualization analysis, and ablation experiments verified that the proposed platform achieved good classification performance with an accuracy of 97.1%. Furthermore, key characteristic peaks for early cervical cancer were successfully captured from the loading plots. This method provided a noninvasive, feasible, and efficient spectral detection strategy for early clinical screening and auxiliary diagnosis of cervical cancer, and exhibited good application potential. It is expected to serve as a promising auxiliary tool for clinical translation of SERS-based cancer diagnosis.