Characterization of trabecular architecture in human femur radiographic images using directional multiresolution transform and AdaBoost model

Abstract

In this work, directional multiresolution curvelet transform is performed in radiographic images to characterize the trabecular structure. The trabecular regions of normal and abnormal human femur bone images are used for the study. The regions of interest such as femoral neck and head are analyzed and compared. The curvelet coefficients are calculated based on each scale and orientation for trabecular images. The mean and energy of the curvelet coefficients associated with each subband are computed. These values are used as the texture feature vector elements to evaluate changes taking place in the trabecular architecture. The three most significant mean and energy feature vector are found using principal component analysis and these values are used as an input to the Adaboost classifier. The results show that the architectural variations are more in the femoral neck when compared to femoral head. AdaBoost classifier performs better in terms of sensitivity (90%) and specificity (100%) for the chosen parameters for femoral neck region when compared to head regions. © 2012 Springer-Verlag.