Multi-Category Bone Fracture Detection Based on Deep Learning in X-ray Imaging Using YOLOv8s

Bone fractures must be accurately detected and classified for optimal patient outcomes and successful treatment. Conventional radiological techniques are prone to fluctuation and require expert interpretation. This work describes a deep learning method for identifying and categorizing five types of fractures in X-ray images: fracture, line, angle, messed-up angle, and background. The technique makes use of the YOLOv8s model. A dataset of annotated X-ray pictures were used to train and assess the model. All 440 image subjects of bone fracture from a publicly available resource (Roboflow) dataset were included in this study. 75% of the images were used to train the deep learning model and 25% were assigned to the testing set. In each class of bone fracture images, 88 images were used to train the proposed model. The results showed that in all five classes, the YOLOv8s model achieved well. The model's accuracy and recall were 93.4% and 90.1%, respectively, while its overall MAP was 91.2%. With MAP scores of 94.5% and 92.7%, respectively, the model demonstrated exceptional efficiency in identifying fractures and lines, according to the performance breakdown by class. Strong performance was demonstrated by the somewhat lower MAP scores for recognizing angles and screwed up angles, which were still above 85%. According to the findings, YOLOv8s is a useful tool for automating the identification and categorization of bone fractures, which could enhance the efficiency and precision of diagnostic procedures in clinical settings.