Various Types Fracture Labeling In Bone Radiographs Using Modified AC-GAN

Abstract

For most of the clinical fracture examinations, orthopedists diagnose symptoms using X-ray images. However, there might still be some details mistaken due to various types of fracture from different parts of arm. Whereas there have been many researches involved in fracture radiographs analysis in attempts to enhance the accuracy of diagnosis, there is still a lack of method to label the position of fracture precisely. Therefore, we proposed using modified Auxiliary Classifier Generative Adversarial Network (AC-GAN) model with adding conditional input to label the fracture surface in different parts of arm. The result shows that the accuracy reaches 91.2%. Comparing with the GAN trained without classification, it turns out that the accuracy reaches only 75.5%. Meanwhile, we also invited two orthopedists to evaluate the results of our model, and it proves that the fracture labeling method we proposed is reliable enough to assist orthopedists.

Ähnliche Arbeiten