Assessment of Different Structures for Predicting Bone Age using Statistical Appearance Models

We compare the utility of models of different structures in the hand for predicting skeletal maturity in young people. Bone age assessment is important for diagnosing and monitoring growth disorders. Statistical models of bone shape and appearance have been shown to be useful for estimating skeletal maturity. In this work we investigate the effect of the choice of region to model on the prediction performance. By analysing the performance on a dataset of 170 digitized radiographs of normal children we show that improvements in accuracy can be achieved by using models of the joint complexes and bones constructed by automatic registration, compared to ones built from manual annotation. We also investigate how sets of joint complexes and bones can best be combined to estimate overall skeletal age. Our results show that the best predictions are obtained from 13 RUS complexes and their combination into eight bone complexes. The mean absolute prediction errors of 0.80 (13 joint complexes) , 0.81(eight bone complexes) years for females and 0.93, 0.94 years for males respectively, demonstrates that similar results are obtained whether we use many local models of individual joints, or a smaller number of models of groups of bones.