Mitral valve endocardiosis is a heart disease that is present in approximately 75% of cases of canine heart disease. Currently, the gold standard for diagnosis is echocardiography, but radiography provides us with valuable information for clinical staging of the disease, such as detecting left atrial enlargement, the presence or absence of pulmonary congestion and cardiogenic pulmonary edema more quickly. The aim of this study was to evaluate the effectiveness of two neural networks, Resnet50v2 and Inception V3, in detecting left atrial enlargement. A database was built with 1052 lateral images. The images were then separated into three distinct groups: 198 images with left atrial enlargement (LAA), 124 images with left atrial enlargement simultaneously with pulmonary edema (LAA_EP), totaling 322 images with left atrial enlargement and 580 images considered normal (N). As a result, all the models were > 90% accurate; Resnet50V2: 94.77%; Inception V3: 92.84%. In conclusion, the use of artificial neural networks to classify left atrial enlargement in radiographs of dogs is a viable approach that can help veterinary radiologists screen patients for left atrial enlargement.