Classification of X-ray images with deep learning algorithms

Nowadays, artificial intelligence algorithms are frequently used in modern healthcare systems to assist staff due to their speed and accuracy. It is important to quickly and accurately diagnose bone fracture problems experienced by individuals living in places where there is no orthopedic services or far away from the health care system. In this study, X-ray images are processed with deep learning algorithms to determine whether there is a fracture or not, and if there is a fracture, to detect the location of the fracture. You Only Look Once (YOLOv8 and YOLOv9) algorithms were used in the study. The evaluations showed that the best results were obtained with the 9th version of the YOLO algorithm, which is known as the most up-to-date algorithm that allows real-time object detection. With the new approach we have developed and with usage of YOLOv9 algorithm, a mAP50 value of 97% was achieved. The MURA v1.1 dataset, which is one of the largest publicly available radiographic image datasets created by the Stanford University Machine Learning Group, was used to develop the deep learning model. This dataset contains a wide range of fracture types and is ideal for training and testing the model. As a result, an application has been developed to quickly and accurately diagnose bone fracture problems experienced by individuals living in areas where access to professional healthcare services is difficult or impossible, and at the same time minimize the diagnostic errors due to overcrowding in emergency services.