A large percentage of the elderly population is suffering from osteoporosis, a disease which heightens the risk for fractures because of the growing loss of bone mass and structure. For diagnosis, Quantitative Computed Tomography (QCT) is used to generate cross-sectional images, allowing a separation of trabecular and cortical bone as well as a precise measurement of the bone mineral density (BMD). Simultaneously, the physician has to determine the trabecular structure of cancellous bone to evaluate the degree of osteoporosis according to a pre-set classification scheme. Since it is difficult to exclude subjective influences using this technique, the R&D team at the Centre for Medical Diagnostic Systems and Visualisation (MeVis) at the University of Bremen has developed an automated procedure to calculate the fractal dimension of the trabecular bone in combination with the BMD values in the cortical shell.
The current structural analysis, based on the visual evaluation of QCT images and grey value profiles to detect the level of osteoporosis, has been computed by the Bremen team to allow for automated classification. The researchers elaborated their method by means of thirty lumbar vertebrae extracted from ten cadavers. They sawed out the mid vertebral slice and x-rayed it on mammography film. This film was scanned to acquire the digital images to work upon. First, they segmented the spongiosa with use of a wavelet based edge detection algorithm in order to remove the corticalis and outer region from the image. The image was binarised at a certain threshold level to obtain the fractal dimension D by means of the box counting method. Variation of the threshold value enabled the team to compute the dimension D in dependency of the treshold value T for each slice.
The average grey value of the respective image normalises the threshold value so that comparison between different densities is possible, in total independence from the bone mineral density. Typically, the lower the threshold values, the larger the selected trabecular area, resulting in an approximately filled plane. More and more parts of the structure disappear when increasing the threshold values, thus decreasing the dimension and exposing the characteristic relationship between the fractal dimension and the threshold value. In addition, the team used QCT to determine the bone mineral density and here, the dimension/threshold function showed the same typical behaviour.
Second, the researchers applied High Resolution Computed Tomography (HR-CT) in order to evaluate the structural changes in the cortical shell. As a matter of fact, osteoporosis causes small, circular cavities of reduced BMD. The areas of low BMD appear in connected regions, called clusters. Using a threshold, these areas can be separated and counted. The higher the threshold values, the larger the areas of the selected cortical bone. Reducing the values provokes disconnection of the clusters, forming new clusters of their own. The lower the threshold values, the more areas become deselected, diminishing the number of clusters even further.
The total number of clusters in dependency of the threshold value shows a local maximum, which is higher in the osteoporotic cases than in the non-osteoporotic ones and therefore constitutes a structural parameter for the corticalis. Thus, the Bremen team succeeds in a fairly precise diagnosis of the patient's level of osteoporosis combining fractal analysis of the trabecular structure with the structural analysis of the cortical shell. Please, check their web site for elucidating images and tables at the Zentrum f¸r Medizinische Diagnosesysteme und Visualisierung, situated at the University of Bremen.