Bone Texture Modeling and Analysis in Radiologocal Diagnosis

Background: Radiologists can diagnose osteoporosis by inspecting X-Ray images of the bone. We therefore assume that changes in the structure of the bone due to the disease should be reflected in changes in both the textural and structural attributes of bone X-Ray images. Previous studies have already shown that the fractional Brownian motion model (fBm), well represent the stochastic attributes of bone texture. The goal of the present study is to refine further the model by incorporating phase information that should account for the structural information that is even more noticeable to the radiologist in his inspection of the images.

Method: We separate the stochastic and structural attributes of the bone images into two distinct layers and first confirm that randomizing the phase of the pure stochastic layer doesn`t affect its subjective appearance and/or its quantified contrast. We then confirm that 1D joint histogram of the wavelet coefficients of the image represented by this layer obey the Gaussian behavior of fractal images. We finally characterize quantitatively the Hurst parameter of the pure fBm component of the image, other moments that account for non-fractal image attributes, and the phase that dominates the other layer.

Results: Preliminary results indicate that the fBm is an important dominant component that characterizes bone structure. It is likewise demonstrated that structural (phase) information should also be used in differentiation between healthy bone and bones that have been affected by osteoporosis.

Conclusion: much more data should be analyzed before automatic diagnosis of osteoporosis will become a reality.