Prediction of Skeletal Age Through Cervical Vertebral Measurements Using Different Machine Learning Regression...

Area of Science:

• Orthodontics
• Radiology
• Biometrics

Background:

• Accurate skeletal age assessment is crucial for orthodontic treatment planning.
• Hand-wrist radiographs (HWRs) are the traditional method, but lateral cephalometric radiographs (LCRs) offer an alternative.
• Regression methods can be used to determine skeletal age from LCRs.

Purpose of the Study:

• To compare skeletal ages derived from LCRs using three regression methods (Ridge, LASSO, ElasticNet) with skeletal age from HWRs.
• To evaluate the explainability power (R2) and consistency of vertebral skeletal age (VSA) against chronologic age (CA) and Greulich-Pyle (GP) age.

Main Methods:

• Analyzed LCRs and HWRs from 794 individuals (7-18 years old).
• Determined hand-wrist skeletal age using the Greulich-Pyle (GP) atlas.
• Calculated VSA using Ridge, LASSO, and ElasticNet regression on 44 linear and 9 angular measurements from cervical vertebrae (C2-C5).

Main Results:

• LASSO regression demonstrated the highest explainability power for VSA (R2: 0.783 for boys, 0.741 for girls).
• Vertebral depth and posterior C3 height showed high beta coefficients in LASSO regression.
• Bland-Altman analysis revealed wider limits of agreement for GP age in boys and for CA-VSA in girls.

Conclusions:

• Despite high R2 values, VSA did not show superiority over chronologic age for skeletal age assessment.
• No significant clinical advantage of VSA over traditional methods was observed.
• For the Turkish population, Greulich-Pyle (GP) age from HWRs is recommended for orthodontic treatment planning.