The thesis will focus on the explainability of machine learning methods in digital pathology, namely on the explainability of neural networks trained for cancer detection in microscopic images of prostate biopsies. The work aims to address the problem of computation and graphical representation of saliency maps during neural network learning. The goal is to understand and evaluate how the network progressively learns feature extractors in cancer tissue classification. The student will implement the solution to the above problem and evaluate it on prostate biopsy images from the Masaryk Cancer Institute in the cloud infrastructure of the RationAI group.