Computational pathology has become a discipline which has widely benefited from the use of data-driven approaches to enable effective prognostics and diagnostics. A number of mathematical and computational models exist aiming for detection and localization of imaging biomarkers that indicate the condition of diseases. In spite of recently achieving great success, there seems to be a clear consensus on the challenges encountered on the use of these methods. The computational tasks to be performed can be error-prone, due to a number of factors such as the high variance of the data or lack of labels. Moreover, the associated models are normally developed for a specific kind of cancer and may not work in other contexts. In this chapter we discuss some of these challenges, and more importantly, we describe which solutions exist to address them.