Determination of the trabecular bone direction from digitised radiographs

There is increasing evidence for monitoring the bone trabecular structure to explain, in part, the mechanical properties of bone. Despite the emergence of Computed Tomography, a radiograph is the standard format as it is cheap and used for assessing implant performance. Furthermore, various image-processing techniques developed to assess the trabecular structure from radiographs have regained interest owing to improvements in imaging equipment. This study assessed the precision and accuracy of the Co-occurrence and Run-length matrix, Spatial-frequency and Minkowski-fractal techniques to infer the trabecular direction from radiographs. Ten clinical images of femoral neck regions were obtained from digitised pelvic radiographs and subsequently analysed. These data were also used to generate synthetic images where the trabecular thickness, separation and directions were controlled in order to calculate the accuracy of the techniques. Additionally, a Laplacian noise was added in order to infer the precision of the techniques. All methods assessed the trabecular direction with a high degree of accuracy in these synthetic images including a single direction and no noise. However, only the Spatial-frequency and Co-occurrence matrix methods performed well on the clinical and heavily corrupted synthetic images. This demonstrated the possibility of inferring a linear trabecular direction in clinical conditions.