An artificial neural network architecture for skeletal age assessment

Skeletal age assessment is a common and time-consuming task in pediatric radiology. In this work we describe a system which implements the TW2 method, using a neural network architecture. Each bone complex is localized on the image, and preprocessed using either a Gabor transform or a multi-scale Difference of Gaussian filtering. The output of the preprocessing stage is fed to a set of neural networks trained to classify each bone accordingly to the TW2 method. Afterward, the skeletal age is estimated and compared with the classification of an expert radiologist.