Non-muscle Invasive Bladder Cancer (NMIBC) accounts for 80% of all bladder cancers. Although it is mostly low-grade tumors, its high recurrence rate necessitates three-times-monthly follow-ups and cystoscopy examinations to detect and prevent its progression. A rapid liquid biopsy-based assay is needed to improve detection and reduce complications from invasive cystoscopy. Here, we present a rapid spectroscopic method to detect the recurrence of NMIBC in urine. Urine samples from previously-diagnosed NMIBC patients (n = 62) were collected during their follow-up visits before cystoscopy examination. Cystoscopy results were recorded (41 cancer-free and 21 recurrence) and attenuated total refraction Fourier transform infrared (ATR-FTIR) spectra were acquired from urine samples using direct application. Spectral processing and normalization were optimized using parameter grid searching. We assessed their technical variability through multivariate analysis and principal component analysis (PCA). We assessed 35 machine learning models on a training set (70%), and the performance was evaluated on a held-out test set (30%). A Regularized Random Forests (RRF) model achieved a 0.92 area under the receiver operating characteristic (AUROC) with 86% sensitivity and 77% specificity. In conclusion, our spectroscopic liquid biopsy approach provides a promising technique for the early identification of NMIBC with a less invasive examination.
Keywords: FTIR; bladder cancer; liquid biopsy; machine learning.