Trabecular bone characterization using circular parametric models

Texture analysis of radiographic bone X-ray images presents a major challenge for pattern recognition and medical applications. Classifying such textures from osteoporotic and healthy subjects is a difficult task. In this paper, we propose a new approach combining wavelet decomposition and parametric circular models to capture the statistical behavior of phase coefficients. We demonstrate that, unlike the magnitude components, the wavelet phase coefficients convey local and structural information across scales and orientations which are of great interest for the study of trabecular bone texture. To assess how well the proposed circular models fit phase coefficients, the statistical test of Kuiper and graphical analysis Quantile-Quantile plots were used. The Support Vector Machine (SVM) and the Neural Network (NN) classifiers were used to evaluate the efficiency of the proposed models to classify two populations composed of osteoporotic patients and control subjects. Using Gabor filters and the Wrapped Cauchy model, an Area Under Curve (AUC) rate of 96.45% was achieved with the SVM classifier. To compare the performance of the proposed parametric approach to other non-parametric texture analysis techniques, the Receiver Operating Characteristic (ROC) analysis was performed. Results have proven that the proposed approach provides the best performance in terms of ROC curves.