摘要：—Skeletal maturity assessment is important for diagnosing and monitoring growth disorders in Children and young adults. Statistical models of bone shape and appearance have been shown to be useful for estimating skeletal maturity. One critical requirement in automated skeletal maturity estimation is matching built models to unseen images of the bones of the hand. Oftentimes some form of initialization is required to prevent the model from falling into local minima. In this work we used Markov Graph Shape models (MGSM) to initialize the image of an incoming radiographic image and then fit a global Active Appearance models of the whole hand using the found points from the Markov Graph Shape models as ’weighted’ constraints. Having found the approximate positions of the bones of hands, we then fit local models to refine the model fit. By analysing performance on dataset of 70 digitized images of normal children we achieved a model fitting accuracy of an overall point-to-point median error of 0.67mm.
