Fracture Detection in Bone: An Approach with Versions of YOLOv4

Radiologists can sometimes overlook fractures because they are difficult to spot. A vast number of studies of deep learning on images have provided interesting applications to medical data analysis, with significant improvements in image-based diagnosis. This study has leveraged YOLOv4 versions that are expected to detect fractures in the wrist bone. The rigorous testing of three levels showed that the YOLOv4-based architectures obtained significantly better results than the state-of-the-art method based on the U-Net model. Our method is evaluated on a public dataset containing over 20,000 X-ray images of wrist fractures to conduct the experiments. The YOLOv4 achieves an accuracy of 0.89871, recall of 0.89871, precision of 0.90369, and F1 of 0.89997, outperforming the U-Net model.