The explainability of neural networks used in domains like medicine is crucial for building trust, ensuring accuracy, uncovering insights, and meeting regulatory needs. It enables healthcare professionals to understand and validate AI decisions, leading to safer and more effective care. The thesis aims to identify, implement and evaluate explainability methods for an existing VGG-16-based neural network trained on histopathological images, focusing on their computational performance. The student will familiarise themself and present existing explainability methods and metrics commonly used for their evaluation. They will discuss and select suitable methods and design, perform and assess the comparative test. As a baseline, the occlusion-based saliency producing semantically valid explanations but being notably slow will be used. The student will optionally contribute the code to HistoPipe, the RationAI open-source library for digital histopathology.