Chest radiography represents an important stage of analysis, allowing the detection of pulmonary opacities, commonly known as infiltrations, which, if detected early, can improve medical follow-up. However, this technique may be limited when the case history under examination represents a complex characterisation of the pathology. Artificial intelligence is proposed as an ideal tool for the analysis of large-scale X-ray images, capable of providing an initial classification of images containing possible infiltrations, providing support to domain experts. The author’s Phd activity focuses on providing artificial intelligence solutions in public, administrative and non-administrative fields, such as the medical sector. The aim is to speed up processes and optimise diagnosis, without replacing a domain expert but providing him with valid tools. In this preliminary study, the results produced by some neural networks are analysed with the focus on understanding the nature of the classifications produced in chest x-rays images field. The aim is to highlight typical problems in the domain and increase the explainability of the results, through the use of the Grad-CAM technique. Possible solutions and future developments of the work are therefore discussed.