Feature selection for diagnozing the osteoporosis by femoral neck X-ray images

We analyzed the quality of a number of features describing the texture of digital X-ray images of the bone tissue for the computer-aided diagnosis of the osteoporosis. We introduced four heuristic features, also considering thirteen adjusted quadratic features described in a previous paper. We solved a problem of selecting the smallest feature subset in order to provide the linear separability of the feature vectors from the learning sample in the corresponding feature space. During the experimental studies we found that the subset of four heuristic features fulfils the separability condition as well as the subset including three quadratic features and one heuristic feature does.