Using Machine Learning and Urine Cytology for Bladder Cancer Prescreening and Patient Stratification

Bladder cancer presents a major and increasing health risk worldwide with early diagnosis being recognized as key to its successful management and treatment. This is the first work to propose the use of data extracted from immunofluorescence images together with readily available data collected from patients referred to cystoscopy as a means of stratification and in particular for the reduction in the number of unnecessary further cystoscopies. We present a thorough analysis of the problem and thus a carefully designed machine learning based solution that achieves this goal. Using a real-world data corpus and thorough statistical analysis, we demonstrate that our method is capable of distinguishing patients at high risk of having bladder cancer with an error rate of less than 5%, who can be prioritized for follow-up examination and treatment, while at the same time reducing unnecessary financial and resource burden, as well as substantial patient discomfort, by correctly identifying 66% of low risk patients with cystoscopy associated morbidities.