Noninvasive detection and monitoring of multiple myeloma by urinary surface-enhanced raman spectroscopy and machine learning

Abstract Abstract Background: Multiple myeloma (MM) is the second most common hematologic malignancy, characterized by high relapse rates and a lack of curative therapies. MM's diagnosis and treatment-efficacy monitoring is still depend on invasive bone-marrow biopsy, immunofixation electrophoresis, and imaging studies now. These methods are time-consuming and lack sensitivity in early disease stages. Urinary surface-enhanced Raman spectroscopy (SERS) combined with machine learning offers a rapid, non-invasive approach for biomarker detection. Objective:To evaluate the feasibility and diagnostic performance of non-invasive detection of MM using urinary SERS combined with a principal component analysis–random forest (PCA–RF) model. Methods: The study was approved by the Ethics Committee of China–Japan Union Hospital of Jilin University (approval No.: 2025-KYYS-036). This single-center, retrospective study in the Department of Hematology recruited 13 patients with newly diagnosed MM and 13 healthy volunteers matched for age and sex. First-morning midstream urine samples were collected upon admission after ≥12 h fasting and prior to any treatment. Each 200 µL sample was mixed with 40 µL silver nanoparticle solution and analyzed on an ATR3100 Raman spectrometer (785 nm, 100 mW), acquiring 100 spectra per subject (3-scan average; total n = 2600). Spectra were baseline-corrected, normalized, then dimensionally reduced via PCA to the top 50 principal components, and finally classified using a random forest model. Receiver operating characteristic (ROC) area under the curve (AUC), accuracy, sensitivity, and specificity were assessed by five-fold cross-validation. Results:The PCA–RF model achieved a mean AUC of 0.975. Five-fold cross-validation yielded an average accuracy of 90.85%, sensitivity of 91.40%, and specificity of 90.30%. Calibration curves demonstrated excellent agreement between predicted probabilities and observed incidences. The primary discriminative peaks were located at approximately 600 cm⁻¹ (aromatic amino acid vibrations) and 1500 cm⁻¹ (nucleoside metabolite vibrations). Conclusions:Urinary SERS combined with the PCA–RF model enables highly accurate, non-invasive detection and monitoring of MM. Its rapid analysis, strong reproducibility, and excellent patient compliance make it well suited for outpatient and bedside use. Future work will involve prospective, multicenter validation and the integration of metabolomic profiles with standard clinical parameters to refine the model and advance its clinical translation. Keywords: multiple myeloma; non-invasive detection; surface-enhanced Raman spectroscopy (SERS); machine learning; random forest; silver nanoparticles