ABSTRACT
Objective To develop mathematical models for skeletal age determination with multiple statistic method based on the correlation between age and the growth of the epiphysis of extremitas sternalis of clavicle in Shanxi adolescents. Methods The 562 Shanxi sternoclavicular joint samples （454 cases of modelling, 108 cases of external verification） were scanned by the thin-section computed tomography. After volume rendering was obtained, indicators such as area of epiphysis, area of metaphysis, longest diameter of epiphysis and longest diameter of metaphysis of both extremitas sternalis of clavicle were collected. Indicators such as the ratio of area of epiphysis to area of metaphysis, and the ratio of longest diameter of epiphysis to longest diameter of metaphysis of both sides were calculated. Then multiple linear regression and random forest discriminant models were used to build mathematical models for age determination of adolescents. Results The obtained indicators exhibited a strong correlation with age （r>0.85）. The multiple linear regression model for males and females （all 4 indicators entering the model） based on the ratio of longest diameter of epiphysis to longest diameter of metaphysis and the ratio of area of epiphysis to area of metaphysis had an internal validation accuracy rate （±1.0 year） of over 92% and 108 cases had an external validation accuracy rate of over 70% （±1.0 year）. The out of bag error rates of random forest discriminant models were less than 2% for people over 18.0 years old （≥18.0 years old） and under 18.0 years old. The external validation accuracy rates of the 108 cases were over 80%. Conclusion The regression and discriminant models established in this study have certain reliability and accuracy and can be used in age determination of Shanxi adolescents.