Detecting pathological processes is a crucial task for medical image processing and analysisThis paper focuses on classifying MRI images of the knee joint into two classes: healthy and showing signs of osteoarthritis. Convolutional neural networks based on the architectures of EfficientNetB5, InceptionResNetV2, MobileNetV2, MobileNet, and VGG16 were used. The research dataset comprises images of healthy and osteoarthritic knees obtained from patients in the Krasnoyarsk Territory, supplemented by publicly available images from the Kaggle repository. Inadequate number of images and class imbalance were addressed by dataset augmentation using various image rotations. The neural network models were compared based on accuracy scores. The best results were achieved by the EfficientNet-B5 (95.55%) and InceptionResNetV2 (94.39%) models. Both models showed a high degree of specificity and sensitivity, which makes them suitable as screening tools. The results obtained indicate the potential of using neural networks to automate the diagnosis of osteoarthritis in knee MRI scans. #CSOC1120