Serum SERS combined with a K-Nearest Neighbor-Graph Attention Network achieved efficient early screening of osteoporosis with an accuracy rate of 93.32% and sensitivity of 93.90%.
Cross-Sectional
Does machine learning-assisted serum SERS technology improve the early screening and prognostic evaluation of osteoporosis compared to traditional machine learning algorithms?
Machine learning-assisted serum SERS technology provides a highly accurate and potentially feasible method for the early clinical screening and prognostic evaluation of osteoporosis.
Early detection, early intervention, early treatment, and timely prognostic monitoring of osteoporosis are crucial for improving patients' quality of life and lifespan. Currently, the diagnostic methods for osteoporosis are expensive, poorly accessible, and radioactive. Surface-enhanced Raman spectroscopy (SERS) has drawn significant attention in medical diagnosis over the past few years. However, prior research has predominantly centered on intricate substrate development or diagnosing different cancers, with comparatively limited studies investigating patients' serum SERS spectra pre- and post-treatment. Through a comparative analysis of the enhancement capabilities of four silver nanoparticle types, we identified that only those synthesized via sodium citrate reduction enabled clear detection of the characteristic peaks in serum samples. Subsequently, by directly integrating the serum SERS spectra obtained via this substrate with the optimized K-Nearest Neighbor-Graph Attention Network, this study established an optical detection approach dedicated to serum component identification and diagnostic model construction. Compared with traditional machine learning algorithms, this method achieves efficient and accurate early screening of osteoporosis (accuracy rate: 93.32%, precision rate: 94.13%, sensitivity: 93.90%, specificity: 93.30%). In addition, the serum SERS spectra of osteoporosis patients prior to and subsequent to treatment were monitored in this research, and potential characteristic peaks as well as metabolites applicable to prognostic assessment were successfully identified. The outcomes indicate that examining the variations in serum SERS spectra across the treatment phase can effectively measure the treatment effect. These results indicate that serum SERS combined with machine learning will provide a potentially feasible method for the clinical early screening and prognostic evaluation for osteoporosis.
Gao et al. (Sat,) conducted a cross-sectional in Osteoporosis. Serum SERS combined with K-Nearest Neighbor-Graph Attention Network vs. Traditional machine learning algorithms was evaluated on Early screening of osteoporosis (accuracy rate). Serum SERS combined with a K-Nearest Neighbor-Graph Attention Network achieved efficient early screening of osteoporosis with an accuracy rate of 93.32% and sensitivity of 93.90%.