An optimized cosine-similarity-loss-function based metric learning neural network enables precise noninvasive diagnosis of bladder cancer through in vitro SERS spectral analysis of clinical urine samples

Key Points

• Noninvasive diagnosis of bladder cancer achieves high accuracy with optimized neural network methods.
• The model recorded a specificity of 92% and sensitivity of 90% in clinical urine samples.
• Analysis utilizes in vitro surface-enhanced Raman scattering (SERS) to evaluate urine samples.
• Highlights the potential for early detection strategies, emphasizing the need for broader validation.