Computer-Assisted Characterization of Prostate Cancer on Magnetic Resonance Imaging

Prostate cancer (PCa) is one of the most prevalent cancers among men. Early diagnosis can improve survival and reduce treatment costs. Current inter-radiologist variability for detection of PCa is high. The use of multi-parametric magnetic resonance imaging (mpMRI) with machine learning algorithms has been investigated both for improving PCa detection and for PCa diagnosis. Widespread clinical implementation of computerassisted PCa lesion characterization remains elusive; critically needed is a model that is validated against a histologic reference standard that is densely sampled in an unbiased fashion. We address this using our technique for highly accurate fusion of mpMRI with whole-mount digitized histology of the surgical specimen. In this thesis, we present models for classification of malignant, benign and confounding tissue and aggressiveness of PCa. Further validation on a larger dataset could enable improved classification performance, improving survival rates and enabling a more personalized treatment plan.