The role of spectral characteristics of urine in bladder cancer diagnostics

• Published in: Scientific reports Vol. 15; no. 1; pp. 29979 - 13
• Main Authors: Velísková, Martina, Masarovičová, Dominika, Waczulíková, Iveta, Kollárik, Boris, Jacko, Juraj, Hunáková, L’uba, Zvarík, Milan
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 15.08.2025; Nature Publishing Group; Nature Portfolio
• Subjects: 631/67/1857; 631/67/589; 692/53/2421; Absorption; Accuracy; Aged; Artificial intelligence; Big Data; Biomarkers; Bladder cancer; Cellular biology; Chromatography; Chromatography, High Pressure Liquid; Creatinine; Emission measurements; Female; Fluorescence; Fluorescence spectroscopy; High-performance liquid chromatography; Humanities and Social Sciences; Humans; Learning algorithms; Liquid chromatography; Machine Learning; Male; Middle Aged; multidisciplinary; Neural networks; Neural Networks, Computer; Regression analysis; Science; Science (multidisciplinary); Spectrometry, Fluorescence - methods; Spectrum analysis; Statistical analysis; Statistical methods; Urinary Bladder Neoplasms - diagnosis; Urinary Bladder Neoplasms - urine; Urine; Urine; Fluorescence; Absorption; Bladder cancer; Chromatography; Machine learning
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-025-15801-3

Finding a non-invasive diagnostic method with sufficient diagnostic power is crucial for early detection of malignant tumor diseases. The main goal of the presented work is to observe changes in the spectral characteristics of urine between patients diagnosed with bladder cancer and control subjects. Data were obtained through fluorescence spectroscopy and high-performance liquid chromatography (HPLC). The data obtained from multiple fluorescence spectra measurements were graphically represented as excitation-emission matrices (EEMs). In both EEMs and chromatograms, statistically significant peaks and areas were identified, which were evaluated using various statistical methods and machine learning techniques (logistic regression, OPLS-DA, convolutional neural networks). The analysis of urine EEMs did not yield satisfactory results; the highest accuracy was achieved using convolutional neural networks, with a maximum accuracy of 72.1% for the training model. Regarding chromatograms, the best results were obtained by applying convolutional neural networks to chromatogram data, achieving an accuracy of 95.3% for the training model. Established methods of data standardization did not improve performance of discrimination models. Our findings highlight the importance of a multi-parametric approach that captures interactions among spectral features, reflecting not only the complexity of cancer but also inter-individual variability among patients. The integration of urinary spectral data with advanced machine learning methods shows potential for improving patient stratification in BC.