Research on the Intelligent Assessment Algorithm of Bone Age Based on Attention Mechanism

Abstract

Bone age is a reliable index to reflect the maturity of physical development, which is of great significance to evaluate the growth and development of children and adolescents, diagnosis and treatment of diseases. Traditional bone age assessment based on artificial has many problems, such as its time-consuming and subjective result, which may lead to great fluctuation of assessment results. Based on the X-ray image of the hand bones, this study proposes an intelligent prediction model of bone age in Deep Learning based on attention mechanism, combined with the traditional methods of bone age interpretation in Deep Learning. In the pre-processing stage, U-Net is used to remove the background of X-ray image of hand bones, and the dense connection network of attention mechanism is used to extract image features, and the mean absolute error function is introduced to improve the accuracy of this model. In the data set of RSNA competition, the mean absolute error of the method proposed in this study is 0.38 ± 0.10 years old, and obtained the best results reported at present.