Fluorescence metabolomics and machine learning models in non-invasive cancer diagnostics

Primárne karty

Fluorescence metabolomics and machine learning models in non-invasive cancer diagnostics

Monika Švecová1 , Ivana Večurkovská1 , Jana Kaťuchová2 , Jana Mašlanková1 , Katarína Dubayová1 , Mária Mareková1
1 Pavol Jozef Šafárik University in Košice, Faculty of Medicine, Department of Medical and Clinical Biochemistry
2 Pavol Jozef Šafárik University in Košice, Faculty of Medicine, 1st Department of Surgery UPJŠ a UNLP

Background and Aim: Colorectal cancer (CRC) remains a leading cause of cancer-related mortality worldwide, largely due to late-stage diagnosis (Oh and Joo, 2019). Early detection significantly improves prognosis, yet current diagnostic methods are often invasive or costly. The incidence and mortality of CRC are notably rising in Slovakia, emphasizing the need for improved diagnostic techniques (Pham et al., 2023). This study explores the potential of spectral fluorescence analysis of urine as a rapid, non-invasive diagnostic tool for CRC, leveraging advanced machine learning techniques to distinguish between control subjects and CRC patients based on their urine's fluorescent characteristics.

Methods: The study involved 206 participants, categorized into control and CRC patients, with the latter including both benign and malignant stages. Urine samples were analyzed using a fluorescence spectrophotometer, measuring in an excitation range between 250 and 550 nm (Dubayová et al., 2023). The fluorescence data were processed and analyzed using GraphPad Prism 8 to evaluate statistical significance and to extract key spectral zones (Z1-Z7) as well as their potential ratios indicative of metabolic variations (Birková et al., 2020). Data was further analyzed using the Python programming language to employ Partial Least Squares Discriminant Analysis (PLS-DA) for the assessment of the clinical application of fluorescent data to potentially diagnose CRC. The machine models were implemented following the guidelines of IFCC (Master et al., 2023).

Results: The PLS-DA model effectively distinguished between control and CRC samples, achieving an initial accuracy of 83 % and an AUC of 75 %. Among the spectral zones, Z1, Z5 and Z6 statistically differentiated between the samples (p < 0.0001). Additionally, the fluorescence intensity ratios Z1/Z2 and Z5/Z6 demonstrated their potential as biomarkers for the early detection of CRC. ROC analysis further validated the diagnostic value of Z5 and Z6 with scores of 74 % and 80 %, respectively. Subsequent optimization of the PLS-DA model, which focused fluorescent zones, substantially improved performance, increasing the accuracy to 90 % and the AUC to 95 %. This enhanced model dramatically reduced false negatives, significantly boosting the sensitivity and specificity required for the early detection of CRC.

Conclusion: Spectral fluorescence analysis of urine presents a promising method for non-invasive CRC screening. This technique's ability to effectively distinguish between healthy individuals and those with CRC highlights its potential integration into routine diagnostics, offering a cost-effective tool for early detection. Future work will focus on refining the diagnostic model and validating these findings in a larger cohort to establish robust clinical guidelines.


This study was supported by the following Scientific Grant Agency of the Ministry of Education, Science, Research and Sport of the Slovak Republic: 1/0435/23.This work has been carried out in collaboration with the 1st Surgical Clinic of the Louis Pauster University Hospital in Košice.


Birková, A. et al. (2020) ‘Human fluorescent profile of urine as a simple tool of mining in data from autofluorescence spectroscopy’, Biomedical Signal Processing and Control, 56, p. 101693. doi: 10.1016/j.bspc.2019.101693.

Dubayová, K. et al. (2023) ‘Visualization of the composition of the urinary fluorescent metabolome. Why is it important to consider initial urine concentration?’, Methods and Applications in Fluorescence, 11(4), p. 045004. doi: 10.1088/2050-6120/ace512.

Master, S.R. et al. (2023) ‘Machine Learning in Laboratory Medicine: Recommendations of the IFCC Working Group’, Clinical Chemistry, 69(7), pp. 690–698. doi: 10.1093/clinchem/hvad055.

Oh, H.-H. and Joo, Y.-E. (2019) ‘Novel biomarkers for the diagnosis and prognosis of colorectal cancer’, Intestinal Research, 18(2), pp. 168–183. doi: 10.5217/ir.2019.00080.

Pham, P.T. et al. (2023) ‘Joinpoint analysis of colorectal cancer trend in the Slovakia’, Bratislavske Lekarske Listy, 124(11), pp. 833–841. doi: 10.4149/BLL_2023_128.