Construction of artificial intelligence-based prediction models for non-recognizable thoracolumbar compression fractures by X-ray inspection / 实用放射学杂志

ABSTRACT

Objective To evaluate the potency of applying an artificial intelligence(AI)based model for classifying vertebral fractures in lumbar X-ray images.Methods Patients who underwent lateral lumbar X-ray and MRI were retrospectively selected.Based on MRI results,the vertebrae were categorized as fresh fractures,old fractures,and normal vertebrae.A ResNet-18 classification model was constructed using delineated region of interest(ROI)on the X-ray images,and the model's performance was evaluated.Results A total of 272 patients(662 vertebrae)were included in this study.The vertebrae were randomly divided into training(n=529)and validation(n=133)sets.The model's performance in discerning normal vertebrae,fresh fractures,and old fractures revealed accuracy of 0.91,0.42,and 0.75,and the sensitivity were 0.91,0.408,and 0.72,while the specificity were 0.796,0.892,and 0.796,respectively.Conclusion The X-ray-based ResNet-18 AI model has significant accuracy for distinguishing old fractures and normal vertebrae;However,the model's accuracy needs further improvement for distinguishing fresh fractures.