Intraoperative guidance plays a pivotal role in enhancing surgical success rates and optimizing patients' prognosis. However, during surgery, the lack of reliable monitoring methods remains a critical challenge. Therefore, we developed an autoencoder-support-vector-machine (SVM)-assisted label-free differential surface-enhanced Raman spectroscopy (dSERS) platform for rapidly intraoperatively assessing parathyroidectomy outcomes. Using only 2 μL of untreated plasma, this platform enables real-time differentiation between complete and partial parathyroid gland resection within 16 min. By leveraging differential spectral analysis (postoperative vs preoperative spectra), our approach effectively minimized individual variability while amplifying surgery-induced molecular changes. The SVM classifier achieved exceptional diagnostic performance, with 95.8% and 79% accuracies in an internal test set and an independent validation cohort (n = 144 and 33 spectra), respectively, suggesting that because of its microliter-scale sample requirements and rapid turnaround time, the label-free dSERS-artificial intelligence platform should become a transformative tool for guiding precision endocrine surgery.
Keywords: differential surface-enhanced Raman spectroscopy; intraoperative guidance; machine learning; parathyroidectomy.