Diagnosis of urogenital cancer combining deep learning algorithms and surface-enhanced Raman spectroscopy based on small extracellular vesicles.

Purpose: To identify and compare the capacities of serum and serum-derived small extracellular vesicles (EV) in diagnosis of common urogenital cancer combining Surface-enhanced Raman spectroscopy (SERS) and Convolutional Neural Networks (CNN).

Materials And Methods: We collected serum samples from 32 patients with prostate cancer (PCa), 33 patients with renal cell cancer (RCC) and 30 patients with bladder cancer (BCa) as well as 35 healthy control (HC), which were thereafter used to enrich extracellular vesicles by ultracentrifuge. Label-free SERS was utilized to collect Raman spectra from serum and matched EV samples. We constructed CNN models to process SERS data for classification of malignant patients and healthy controls (HCs).

Results: We collected 650 and 1206 spectra from serum and serum-derived EV, respectively. CNN models of EV spectra revealed high testing accuracies of 79.3%, 78.7% and 74.2% in diagnosis of PCa, RCC and BCa, respectively. In comparison, serum SERS-based CNN model had testing accuracies of 73.0%, 71.1%, 69.2% in PCa, RCC and BCa, respectively. Moreover, CNN models based on EV SERS data show significantly higher diagnostic capacities than matched serum CNN models with the area under curve (AUC) of 0.80, 0.88 and 0.74 in diagnosis of PCa, RCC and BCa, respectively.

Conclusion: Deep learning-based SERS analysis of EV has great potentials in diagnosis of urologic cancer outperforming serum SERS analysis, providing a novel tool in cancer screening.