Abstract
This work provides the task of diagnosing osteoporosis using X-ray images of the knee joint using deep learning methods. The purpose of this study is to analyze the effectiveness of various convolutional neural networks for classifying medical images into three main classes.: norm, osteopenia and osteoporosis. The VGG16, ResNet50, EfficientNet-B0, and YOLOv8-cls architectures were used as test models. A hierarchical approach based on step-by-step classification is also applied. These models were trained and evaluated on a single dataset containing 293 test images. This ensured the correctness of this comparative analysis. The accuracy, precision, recall, and F1-score metrics were used to evaluate classification quality and error matrices were analyzed. The results of these experiments showed that the ResNet50 model showed the best ratio of accuracy and stability, reaching an accuracy of 82.3%. But there was confusion between the two classes with similar characteristics. The hierarchical approach has reduced the inter-class confusion between osteopenia and osteoporosis. But it led to a decrease in the overall classification accuracy. The experimental results confirm the effectiveness of using convolutional neural networks for the diagnosis of osteoporosis.