Evaluation of the diagnostic potential of Fourier transform-infrared spectroscopy on urine for urothelial bladder cancer: an in-hospital field study

Abstract

The increasing global incidence of bladder cancer necessitates better diagnostic methods. This study investigates the potential of mid-infrared spectroscopy on fresh urine samples as a non-invasive approach for diagnosing bladder urothelial carcinoma. In order to position our approach as close as possible to real clinical practice, urine samples were collected from patients undergoing cystoscopy in a hospital urology department. The spectral data were analysed using principal component analysis (PCA) and uniform manifold approximation and projection (UMAP). Unsupervised methods did not reveal clear differences between cancerous and non-cancerous samples, supervised models, including support vector machines (SVM) and random forest (RF), were applied to classify patients into cancer and control groups. These models achieved a diagnostic accuracy of approximately 65 %, with a sensitivity of 87 % for high-grade tumours and 100 % for pT2 tumours. Despite these promising results, the overall accuracy remains insufficient for routine clinical use. The inherent variability in urine composition, influenced by factors such as diet and medications, poses challenges in identifying reliable spectroscopic markers. Nonetheless, mid-infrared spectroscopy shows a promising, non-invasive diagnostic approach for bladder cancer. Further research is essential to enhance prediction models and meet the criteria for potential clinical deployment.