Characterization of bone quality using computer-extracted radiographic features

Both bone mineral density (BMD) and trabecular structure are important dete rminates of bone mechanical properties. However, neither BMD or trabecular structural features can completely explain the variations in bone mechanica l properties. In this study, we combine BMD and bone structural features to characterize bone mechanical behavior. Radiographs were obtained from 34 f emoral neck specimens excised during total hip arthroplasties. Each neck ra diograph was digitized and a region of interest (ROI) was selected from the medial side of the femoral neck. Textural features, the global Minkoswski dimension and trabecular orientation, were extracted from each ROI image us ing Minkowski dimension analysis. The BMD of each specimen was measured usi ng dual-energy x-ray absorptiometry (DXA) and subsequently normalized by bo ne size as measured from a standard pelvis radiograph. Mechanical testing w as performed on the trabecular bone cubes machined from each femoral neck t o yield bone mechanical properties. Multiple regression was performed to se lect the best features to predict bone mechanical properties. The results s uggest that, using multiple predictors including normalized BMD structural features, and patient age, the coefficients of determination (R-2) improved over the use of BMD alone. For bone strength, the R-2 was improved from 0. 24 using conventional BMD to 0.48 using a four-predictor model. Similar res ults were obtained in the prediction of Young's modulus, i.e., the R-2 was improved from 0.25 to 0.55 in going from the model using conventional BMD t o a four-predictor model. This study demonstrates the contributions of norm alized BMD, structural features, and age to bone mechanical properties, and suggests a potential method for the noninvasive evaluation of bone mechani cal properties. (C) 1999 American Association of Physicists in Medicine.