Computerized classification of maturity stages of hand bones of children and juveniles

Bone age assessment is an important diagnostic task in pediatric radiology. Most of the current methods depend on a subjective rating of bones in hand radiographs, e.g. the TW2 method of Tanner and Whitehouse (1983). Since these approaches are time consuming and characterized by considerable variability an objective and automatic procedure is desirable. The authors´ developed approach depends on a pattern recognition concept. Typical regions of interest (ROIs) of the X-rays, where the growing process of the bones becomes visible, are used as feature matrices. Mainly the spectral representation are correlated with prototypes corresponding to different maturity stages. A set of 280 hand radiographs are used to evaluate the authors´ procedures. 65% to 80% of the bones under consideration are correctly classified. In less than 7% the mis-classification is more than one stage. Further work is done to improve the differentiation of neighbouring stages