Developing non-invasive bladder cancer screening methodology through potentiometric multisensor urine analysis

Abstract

We report on the feasibility study exploring the potential of a simple electrochemical multisensor system as a tool for distinguishing between urine samples from patients with confirmed bladder cancer (36 samples) and healthy volunteers (51 samples). The potentiometric sensor responses obtained in urine samples were employed as the input data for various machine learning classification algorithms (logistic regression, random forest, extreme gradient boosting classifier, support vector machine, and voting classifier). The performance metrics of the classifiers were evaluated via Monte-Carlo cross-validation. The best model combining all the acquired data from the people aged 19–88 with different tumor grades and malignancies, including patients with recurrent bladder cancer, yielded 72% accuracy, 71% sensitivity, and 58% specificity. It was found that these metrics can be improved to 76% accuracy, 80% sensitivity, and 75% specificity when only a limited age group (50–88 years of age) is considered. Taking into account the simplicity of the proposed screening method, this technique appears to be a promising tool for further research.