Application of the YOLOv5 Neural Network for Pattern Recognition on Radiographic Images of the Human Wrist Joint

The possibility of using artificial intelligence in the task of recognizing objects of interest in medical radiographic images has been studied. To implement computer vision, the YOLOv5 neural network and the use of its functionality for pattern recognition on radiographic images of the human wrist joint were considered. YOLOv5 is one of the most popular neural network models due to a number of advantages and features: high learning speed, object detection accuracy, efficient use of hardware and support for various architectures. The use of an artificial neural network in medicine is a promising direction that allows one to obtain data on pathologies and anomalies on radiographs, improving the diagnostic process and providing quick results with a high percentage of reliability. To adapt the model for the task of recognizing objects of the wrist joint in radiographic images, a data set was labeled, including images for retraining the neural network model and test images. Learning key numerical properties is key to building and tuning effective neural network models. The correctness of their selection and debugging have a direct impact on achieving the desired results when retraining and using neural networks. According to the results of the neural network, all classes of objects were correctly identified in more than 70% of the images; this indicator is a worthy result of the work, which confirms the relevance of using a computer vision tool in medical diagnostics and practical application in clinical practice.