Abstract: Objective  To collect Raman spectral data of bladder tumor tissues and normal mucosa in vivo by using a high-throughput fiber-optic Raman probe, construct a diagnostic model based on Principal Component Analysis Neural Network (PCANet), evaluate its performance, and explore its application in non-invasive real-time diagnosis of bladder tumors.Methods  During cystoscopy, patients undergoing random biopsy or transurethral resection of bladder tumors were included. In vivo Raman spectra were collected from suspicious tumor tissues and normal mucosa by using a high-throughput fiber-optic Raman probe, and tissue samples from corresponding sites were obtained. All sampling sites were confirmed by postoperative pathological examination. Based on pathological results, Raman spectral data were categorized into tumor and normal mucosa groups, and a PCANet-based diagnostic model for bladder tumors was constructed.Results  A total of 150 patients were enrolled, among whom 134 were pathologically diagnosed with bladder tumors postoperatively. In vivo Raman spectra were collected from 196 tumor tissue sites and 329 normal mucosa sites. Significant differences were observed at six characteristic peaks: 916, 1 094, 1 290, 1 346, 1 436, 1 672 cm-1. The PCANet-based diagnostic model demonstrated a sensitivity of 90.6% and a specificity of 91.3% in classifying in vivo bladder Raman spectra.Conclusion  This study demonstrates that the real-time Raman spectroscopy diagnosis technology based on high-throughput fiber Raman probes can achieve rapid and non-invasive in vivo diagnosis of bladder tumors, with excellent diagnostic performance and significant clinical application prospects.

Key words: Raman spectroscopy, non-invasive diagnosis, bladder tumor, neural networks, cystoscopy, fiber optic technology