Classification of bone pathologies with finite discrete shearlet transform based shape descriptors

Bone edema is a nonspecific and reactive condition of bone which is easily detectable with PD weighted MRI. In this study we decomposed segmented PD weighted MR images of humeral head, based on finite discrete shearlet transform (FDST) which provides optimal multiscale and multidirectional representation of 2D signals. Afterwards shape features were extracted from coefficients of FDST based on Pyramid of Histograms of Orientation Gradients (PHOG) method which captures the local image shape and its spatial layout. Next we classified extracted humeral bone features as edematous and normal with support vector machine (SVM). We compared the success rates of classification of PHOG and FDST based PHOG features. Experiments delivered highly successful classification results with FDST based PHOG descriptors than PHOG features alone. Our proposed method is promising for automatic diagnosis of humeral head artifacts.