A variational model for trabecular bone radiograph characterization

Abstract

This paper deals with texture characterization of trabecular bone radiographs. The proposed method is based on the variational decomposition of the original image as three components which describe the macro- and the micro- texture as well as a third part representing contrast and luminance of the image. Each component is analyzed using different texture characterization methods. The extracted texture features are used to separate between two different populations composed of osteoporotic patients with various bone fractures and control subjects. Results show that depending on the component used, separation rate varies. Especially, the type of fracture and the statistical power of separation were improved using the micro texture component. Classification rates were improved from 48% and 66% to 85%.