Trabecular Pattern Analysis of Proximal Femur Radiographs for Osteoporosis Detection

The most common approach to the study of the structural changes with age in the proximal femur is by means of the Singh index. However, subjectivity of its grades determination may limit its practical value. In this paper, we used several texture analysis techniques to characterize the trabecular pattern changes in proximal femur recorded in radiographs that lead to osteoporosis detection. Texture analysis was used to rank 41 radiographs and then compared to Singh index grading performed by experienced experts. We compare three texture analysis methods namely Gabor filter, wavelet transforms and fractal dimension to extract the features that represent the structural change in trabecular pattern. The assessment of osteoporosis can be performed by observing and analyzing the trabecular patterns recorded in proximal femur radiographs. The predetermined Singh indexes of trabecular pattern are used to justify the classification result. The features extracted using Gabor filter, wavelet transform and fractal analysis are found to be significantly correlated with predetermined Singh indexes (r=0.87, r=0.84 and r=0.79 with p=0.001).