Non-alcoholic fatty liver disease (NAFLD) has shown an increase in prevalence worldwide in recent decades, which has led to several investigations focused on its early and non-invasive detection. In countries such as Argentina, Chile, and Spain, multiple image-based diagnostic methods have been developed to improve accuracy. In contrast, Mexico still requires further technological development to address this condition with advanced computational tools. The aim of this work is to implement the use of convolutional neural networks for the classification of medical images based on patterns present in patients with hepatic steatosis. To this end, a computational model was designed and trained with clinical studies of both positive and negative NAFLD cases for the automatic detection of fatty liver.