Analysis of Modern Image Classification Platforms for Bone Fracture Detection
Adisorn Kheaksong, Parinya Sanguansat, Pongsakorn Samothai, Tinnaphob Dindam, Kanabadee Srisomboon, Wilaiporn Lee
Abstract
Convolutional Neural Network (CNN) is a powerful and widely used technique for applications that want to extract feature images such as face recognition or object detection. The detection of bone fractures is one of the most interesting application for medical treatment. The detection of bone fractures helps in medical treatment to reduce the diagnosis error from human limitations. YOLO is an interesting CNN model used in fracture detection, but YOLO-R, YOLO-X and YOLOv7 have not been assessed on X-ray images. Therefore, the efficacy of YOLO-R, YOLO-X and YOLOv7 was studied under long bone x-rays to determine the location of fractures. The results showed that YOLO-X outperformed YOLO-R and YOLOv7 due to higher accuracy and better convergence time. Since the YOLO-X model located the fracture area with alternative processing such as detection head decoupling, anchors-free and augmentation strategies. Then, it can detect the fracture locations even if the features of X-ray images are low.