Machine learning-assisted surface-enhanced Raman spectroscopy technique in dried human blood serum analysis for rapid diagnosis of papillary thyroid carcinoma and central lymph node metastasis.

Central lymph nodes are the most frequent metastatic site in papillary thyroid carcinoma (PTC). Prophylactic central neck dissection carries nontrivial risks (e.g., recurrent laryngeal nerve injury and lymphatic leakage), creating a need for less invasive preoperative prediction of central lymph node metastasis (CLNM) status. In this study, serum samples were collected from 150 healthy controls (HC) and 150 patients with PTC, including 75 CLNM and 75 central lymph node metastasis-negative (NCLNM). Surface-enhanced Raman spectroscopy (SERS) spectra were analyzed using multiple machine-learning (ML) models to classify PTC and predict CLNM. Support Vector Machine (SVM) achieved the highest accuracy for HC vs. PTC (99.76%), while Random Forest (RF) performed best for CLNM vs. NCLNM (95.91%). To interpret these models, SHapley Additive exPlanations (SHAP)-based interpretability analysis identified spectral patterns that consistently contributed to model predictions. These results support serum-based, minimally invasive prediction of central lymph node metastasis in PTC and may assist preoperative risk stratification.