An improved AlexNet model for automated skeletal maturity assessment using hand X-ray images

Abstract There are several problems with traditional artificial skeletal age assessment methods, such as strong subjectivity, large random errors, complex assessment processes, and long assessment cycles. In this study, automated skeletal bone age assessment based on deep learning were performed. Other than examining all the bones in the whole hand, we have proposed a skeletal maturity assessment method based on the Standards of Skeletal Development of Hand and Wrist for Chinese (CHN) method that observing only 14 representative hand bones using the deep convolutional neural network (CNN). The method was compared with the traditional method using whole hand evaluated using the same test dataset. We have also expanded the dataset and increase the generalisation ability of the CNN using data augmentation. As a result, this method is able to improve the accuracy of the final skeletal age assessment and reduce the upper limit of the absolute value of the single skeletal age error. The experiments demonstrate the effectiveness of the proposed method, which can provide physicians with more stable, efficient, and convenient diagnostic assistance and decision support.