Automatic quantitative analysis of lumbar bone radiographs

In current radiological diagnostic practice the radiographs often appear as `rough´ means, difficult to understand due to the bad quality of the radiographic images. The aim of the work reported here is to guarantee a diagnostic support even in the presence of radiographic means only, with a little cost increment due to a computer system, by supplying a useful primary screening tool that automatically analyzes digitized radiographs and gives important features relevant from a medical standpoint. The specific domain of image analysis concerns osteoporosis which is a long term disease requiring an accurate measurement of bone density and a periodic follow-up. The automatic image processing method described here goes through a series of steps: selection of the region of interest, first preprocessing, extraction of the vertebral left side, second preprocessing, extraction of upper and lower borders of the end plates, detection of the greatest uniform area inside the vertebral body, third preprocessing, correction of image nonuniformity, detection of the end plates, feature detection (variance, mean-gray level, concavity, body and tissue density, etc.). The automation of the process is complete and the results are in good agreement with those obtained from human analysis