The aim is to investigate clustering-based techniques for size/complexity reduction of neural networks used in computer vision to make them explainable. We consider a concrete problem, cancer detection in whole slide images, on concrete data coming from Masaryk Memorial Cancer Insitute, and, for a start, a concrete network VGG-16. We have already trained the network to the state-of-the-art level of precision on the images from pathology. However, to obtain a production-ready solution, we need to improve the explainability of the whole system substantially. Thus, the next step is to extract information from the network parameters in a humanly understandable form. The selected approach is based on applying clustering techniques to values of tensors that appear during the inference of the network. In particular, to the values of feature maps. This is precisely the goal of the thesis. The work involves exploring the relevant literature on clustering techniques applied to neural networks, implementation of selected clustering techniques, experimentation with the techniques, and integration of the best ones into our framework Pathopus.