Retinopathy of Prematurity (ROP) is an eye disease of premature infants, which in severe cases can lead to complete loss of vision. Timely diagnosis and treatment could help preserve the vision. This study presents a method to automate the diagnosis of ROP using EfficientNetV2B1 and YOLOv8n convolutional neural networks. Experiments were conducted on a set of digital fundus images of premature infants obtained at the regional center for retinopathy of prematurity of the SAHCI SR «Bonum MCMC». Augmentations were used on the training dataset to deal with the lack of data and class imbalance. The Grad-CAM method was used to interpret the features extracted by the EfficientNet model. Experimental results showed that the proportion of correctly classified images was 90% for the EfficientNet model and 95% for YOLO, with YOLO performing better in terms of accuracy and completeness than EfficientNet. Also, the YOLOv8n model did not produce a single false negative, i.e., all images with “Pathology” were correctly identified.