Bladder cancer diagnostics: A digital slide into the future

Up to 78% of patients with non-muscle invasive bladder cancer will develop a recurrence within 5-years after initial diagnosis. Therefore, lifelong follow-up with cystoscopy and upper tract imaging is recommended by the guidelines. Adequate risk assessment is extremely difficult and current risk tools fail to correctly predict the chance of recurrence for patients with non-muscle invasive bladder cancer. The current risk stratification systems are based on clinical and pathological tumor characteristics; however, these parameters are operator dependent and prone to interobserver variability. The histological assessment is an important parameter in the risk stratification, however, the agreement between pathologists for the assessment of aggressiveness (grading) and tumor ingrowth (staging of the tumor are only up to 50% and 60%, respectively.
The objective of this thesis is to develop methods to improve the risk stratification of patients with non-muscle invasive bladder cancer (NMIBC). We therefore investigated the feasibility of reconstructing bladder tumors in 3D out of two-dimensional (2D) histological slides, and whether this can be combined with three-dimensional (3D) mass spectrometry imaging data. We assessed the association of intravesical tumor location on one- and fiveyear recurrence-free survival (RFS) in patients with NMIBC. Finally, we developed a deep learning network to automatically detect and grade urothelial cell carcinoma on histological slides and to predict the one- and five-year RFS in these patients.